Unraveling the molecular Nexus of Alzheimer’s Disease and HIV encephalitis: The role of cellular senescence and transcriptional regulation Authors: Hao Zhang, WenJun Chen, ShuYou Yuan, ShaoXiang Ding, HongXia Bao, Bo Cai, JunKai Sun, Wei Lu, HaoGang Zhu, GuoXun Shi, Eliseo A Eugenin, Eliseo A Eugenin, Eliseo A Eugenin, Eliseo A Eugenin, Eliseo A Eugenin Published: 2026 ABSTRACT Alzheimer’s disease (AD) and HIV-associated neurocognitive disorder (HAND) share progressive cognitive decline. Their common molecular mechanisms remain poorly understood. Current therapeutic approaches lack effective biomarkers for early diagnosis and intervention. We integrated transcriptomic profiles from multiple independent cohorts across brain tissues and blood. We systematically evaluated diagnostic performance using machine learning algorithms including Random Forest, Support Vector Machine, and XGBoost. Notably, FOXO3 emerged as the top cross-disease biomarker. FOXO3 achieved diagnostic accuracy in AD temporal cortex (area under the curve [AUC] = 0.922, 95% CI: 0.885–0.959). FOXO3 showed diagnostic performance in HAND frontal cortex (AUC = 0.771, 95% CI: 0.724–0.818). FOXO3 expression positively correlated with APP (R = 0.558). FOXO3 expression negatively correlated with MAPT (R = −0.690) and SORL1 (R = −0.856). ACE showed strong positive correlation with FOXO3 (R = 0.685), suggesting vascular involvement. STAT3 and ZNF341 were identified from 21 transcription factor candidates. STAT3 ranked first in HAND integrated cohort (n = 107, AUC = 0.759). STAT3 may function as an inflammatory mediator through JAK-STAT signaling. ZNF341 showed strongest transcriptional association with disease status (β = 5.09 in AD, β = 4.85 in HAND). The two-gene panel (FOXO3-ZNF341) achieved diagnostic accuracy in blood samples (AUC = 0.764, n = 329). This performance approaches clinical utility thresholds. Blood-based detection offers non-invasive diagnostic potential. Saturation analysis identified three molecules as optimal panel size. Marginal AUC gains declined below 0.02 beyond this threshold. FOXO3, STAT3, and ZNF341 showed stable selection frequency (97%, 65%, and 100%, respectively). PI3K-Akt and FoxO signaling pathways were enriched, which are known to regulate apoptosis and cell survival. Taken together, these computational findings indicate FOXO3 transcriptional regulatory activity in neurodegeneration. The three-molecule panel represents candidate blood-based diagnostic biomarkers. These candidate biomarkers warrant further investigation in independent clinical cohorts. Introduction Alzheimer’s disease (AD) represents the most common neurodegenerative disorder worldwide. AD affects over 50 million individuals globally and imposes substantial healthcare burdens. The disease manifests as progressive cognitive decline and functional impairment [1]. Pathological hallmarks include amyloid-β plaques and tau tangles. Current therapeutic approaches demonstrate limited efficacy in disease modification [2]. HIV-associated neurocognitive disorder (HAND) constitutes a distinct neurodegenerative condition affecting 30–50% of people living with HIV despite antiretroviral therapy (ART) [3]. Both AD and HAND share overlapping clinical features including progressive cognitive deterioration and quality of life decline [4]. Healthcare costs and caregiver burdens remain substantial in both conditions. These similarities suggest potential common molecular mechanisms underlying neurodegeneration [4]. Current diagnostic approaches rely heavily on clinical assessments. AD diagnosis often occurs after significant neuronal damage has accumulated [5]. HAND diagnosis depends on neuropsychological testing without validated molecular biomarkers [6]. Cellular senescence has emerged as a critical mechanism in neurodegeneration. Senescent cells accumulate in aging brains and secrete inflammatory factors. This senescence-associated secretory phenotype (SASP) promotes chronic inflammation [7]. Recent evidence implicates cellular senescence in both AD and HAND pathogenesis [8,9]. We hypothesized that senescence-related genes drive shared transcriptional alterations. These genes may serve as diagnostic biomarkers for both conditions. Transcriptomic analyses combined with machine learning offer robust approaches for biomarker discovery. Studies identified dysregulated genes involved in inflammation and synaptic function in AD [10]. HAND transcriptomic studies showed immune activation and neuronal dysfunction [11]. However, direct comparisons between AD and HAND remain limited. Machine learning approaches handle high-dimensional genomic data effectively. Ensemble strategies combining multiple algorithms improve diagnostic accuracy. Multi-dataset integration addresses sample size limitations through batch effect correction. Such approaches have identified robust biomarkers in cancer and neurodegeneration. However, systematic comparisons of ensemble methods in AD and HAND remain unexplored [12]. This study employed comprehensive bioinformatics approaches to identify shared molecular signatures. We integrated multiple transcriptomic datasets from NCBI Gene Expression Omnibus (GEO). Machine learning models including Random Forest, Support Vector Machine, and eXtreme Gradient Boosting (XGBoost) identified key molecular markers. We systematically evaluated 129 ensemble algorithm combinations for optimal diagnostic performance. Transcription factor networks were reconstructed using regulatory inference. Our objective was to uncover shared mechanisms between AD and HAND. The findings may inform diagnostic and therapeutic development for both neurodegenerative conditions. Materials and methods Study design and data availability This study analyzed publicly available de-identified transcriptomic datasets from NCBI GEO database. No ethical approval or patient consent was required as per institutional guidelines for secondary data analysis. All datasets are summarized in Table 1. For clarity, dataset information appears in both the summary and Methods sections: the summary provides an overview of dataset types and sources, while the Methods section details specific sample sizes, preprocessing steps, and analytical parameters for reproducibility. 10.1371/journal.pone.0355548.t001Table 1Comprehensive overview of Alzheimer’s and HIV-related neurodegenerative datasets.GSE IDCondition (N)AgeSex (M/F)Diagnosis (per-group N)Disease StageBrain Region/ TissueGSE3489HAND (14) [†]52.3 ± 8.7[§]N/SHIVE+ (8), HIVE- (6)N/SFrontal cortex (BA46)GSE28160HAND (21)N/SN/SART+ HAND (7), ART- HAND (8), Ctrl (6)N/SMultiple brain regionsGSE35864HAND (24) [‡]N/SN/SCtrl (6), HIV (6), HAND (7), HIVE (5)N/SBasal ganglia, White matter, Frontal cortexGSE37263AD (16)78.5 ± 9.2N/SAD (8), Ctrl (8)N/STemporal cortex (BA22)GSE48350AD (84)20–99 yrs[¶]124M/ 129FAD (28), Ctrl (56)N/SHippocampus, Entorhinal cortex, Superior frontal gyrus, Post-central gyrusGSE63060AD (Blood) (329)74.2 ± 7.8191M/ 138FAD (145) [**], MCI (80) [**], Ctrl (104) [**]AD/MCI vs. CtrlBloodGSE122063AD (31)N/SN/SVaD (8), AD (12), Ctrl (11)VaD vs. ADFrontal cortex, Temporal cortexGSE132903AD (195)N/SN/SAD (97), Ctrl (98)AD vs. CtrlMiddle temporal gyrus (BA21)† GSE3489: 14 unique individuals (6 HIVE- + 8 HIVE+), each with A/B technical replicates = 28 arrays total.‡ GSE35864: 24 subjects × 3 brain regions (basal ganglia, white matter, frontal cortex) = 72 samples total.§ GSE3489 age (52.3 ± 8.7 yrs) derived from original manuscript Methods section; age/sex data for other HAND datasets not available in GEO records.¶ GSE48350 spans an age range of 20–99 years, reflecting the heterogeneous donor population from the Harvard Brain Tissue Resource Center.** GSE63060 per-group N from raw GEO data (145 AD, 80 MCI, 104 Control). The AddNeuroMed study final classification reported AD (165) and Ctrl (164) after diagnostic adjudication of MCI cases; see Methods for details.Abbreviations: AD, Alzheimer’s Disease; HAND, HIV-Associated Neurocognitive Disorder; HIVE, HIV Encephalitis; ART, Antiretroviral Therapy; MCI, Mild Cognitive Impairment; VaD, Vascular Dementia; Ctrl, Control; N/S, Not Specified; BA, Brodmann Area. Comprehensive overview of Alzheimer’s and HIV-related neurodegenerative datasets. † GSE3489: 14 unique individuals (6 HIVE- + 8 HIVE+), each with A/B technical replicates = 28 arrays total. ‡ GSE35864: 24 subjects × 3 brain regions (basal ganglia, white matter, frontal cortex) = 72 samples total. § GSE3489 age (52.3 ± 8.7 yrs) derived from original manuscript Methods section; age/sex data for other HAND datasets not available in GEO records. ¶ GSE48350 spans an age range of 20–99 years, reflecting the heterogeneous donor population from the Harvard Brain Tissue Resource Center. ** GSE63060 per-group N from raw GEO data (145 AD, 80 MCI, 104 Control). The AddNeuroMed study final classification reported AD (165) and Ctrl (164) after diagnostic adjudication of MCI cases; see Methods for details. Abbreviations: AD, Alzheimer’s Disease; HAND, HIV-Associated Neurocognitive Disorder; HIVE, HIV Encephalitis; ART, Antiretroviral Therapy; MCI, Mild Cognitive Impairment; VaD, Vascular Dementia; Ctrl, Control; N/S, Not Specified; BA, Brodmann Area. This study integrated multiple transcriptomic datasets to investigate molecular mechanisms underlying AD and HAND. The analysis workflow is illustrated in Fig 1, which outlines the sequential steps from data acquisition, preprocessing, differential expression analysis, machine learning modeling, transcription factor inference, to ensemble diagnostic evaluation. We analyzed GSE37263 (AD temporal cortex) and GSE3489 (HAND frontal cortex). Additional datasets included GSE35864 (HIV brain regions), GSE48350 (AD hippocampus), GSE122063 (AD cortex), and GSE63060 (AD blood). Integrated cohorts included GSE28160/GSE35864 (HAND, n = 107) and GSE118553/GSE132903 (AD, n = 596). Study design and analytical workflow. The seven-step pipeline includes: (i) data acquisition from AD (GSE37263, GSE48350, GSE63060, GSE122063, GSE118553, GSE132903) and HAND (GSE3489, GSE35864, GSE28160) cohorts; (ii) quality control and batch effect correction; (iii) differential expression and enrichment analysis (limma, BH-adjusted p < 0.05); (iv) machine learning biomarker identification (RF, SVM, XGBoost, 5-fold CV); (v) transcription factor inference (Lasso, Elastic-Net, Adaptive Lasso on 21 TFs); (vi) multi-algorithm ensemble evaluation (129 combinations); and (vii) blood-based validation in GSE63060 (n = 329). Data preprocessing and quality control Data retrieval was performed using R version 4.2.1 with GEOquery package. Normalization employed limma package’s normalizeBetweenArrays function. Probes with multiple gene correspondences were excluded, retaining probes with highest signal intensity. For GSE63060, we removed negative values and applied log2 transformation followed by quantile normalization. Two independent dataset pairs underwent batch effect correction: GSE28160/GSE35864 (HAND) and GSE118553/GSE132903 (AD). Principal component analysis (PCA) quantified batch-driven variance in uncorrected data. We calculated variance percentages for each principal component. PCA was repeated after correction to assess integration efficacy. Uniform Manifold Approximation and Projection (UMAP) visualized pre-correction separation. Sample correlation heatmaps evaluated expression consistency. Gene variance distribution on log10 scale confirmed preserved biological variability. For integrated cohorts (HAND n = 107, AD n = 596), data preprocessing used R version 4.5.2. Zero-variance features were removed. Missing values underwent median imputation. Samples received group normalization. Seven core molecular markers were selected for downstream analysis: FOXO3, ACE, STAT3, ZNF341, APP, GFAP, and FOS. Differential expression and functional analysis Differential expression analysis employed limma package with empirical Bayes moderation. Benjamini-Hochberg method controlled false discovery rate. Adjusted p-values < 0.05 were considered statistically significant. Volcano plots visualized fold-change versus statistical significance. We incorporated 279 cellular senescence-related genes (retrieved from GeneCards database, S1 Table) to identify overlaps with differentially expressed genes (DEGs). Enrichment analysis used clusterProfiler R package. We identified overrepresented Gene Ontology (GO) terms and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways associated with FOXO3-linked DEGs. GOplot package calculated z-scores for enriched terms. Gene Set Enrichment Analysis (GSEA) was conducted on GSE37263 dataset to explore AD-associated pathways. GO/KEGG enrichment analysis examined intersecting DEGs from both datasets. Co-expression analysis used Spearman’s rank correlation coefficient to account for non-normal distributions. Correlation coefficients with absolute values > 0.5 and p < 0.05 were considered significant. Correlation networks were visualized using heatmaps with hierarchical clustering (complete linkage method). Z-score normalization standardized expression levels. For small sample sizes, Leave-One-Out Cross-Validation (LOOCV) was employed. Model performance was assessed using root mean square error (RMSE), R-squared, and mean absolute error (MAE). Machine learning framework All statistical analyses used R version 4.5.2. Data normality was assessed using Shapiro-Wilk test. Homogeneity of variance was evaluated using Levene’s test. For non-normally distributed data, non-parametric tests were applied. All tests were two-tailed with significance threshold of α = 0.05. We implemented three primary algorithms: (1) Random Forest with 500 trees (mtry parameter optimized via cross-validation based on previous neurodegenerative studies showing stable performance); (2) Support Vector Machine with radial basis function kernel (gamma and cost parameters tuned using grid search with ranges determined through preliminary experiments); and (3) XGBoost using gradient boosting framework (learning rate = 0.1 balancing training speed and convergence, with 100–500 estimators). Feature importance was assessed using mean decrease in accuracy for Random Forest and gain-based importance for XGBoost models. Recursive feature elimination with cross-validation determined optimal feature subsets, with saturation threshold of 0.02 AUC gain based on previous biomarker panel studies. For transcription factor identification, we focused on 21 transcription factors predicted by ChipBase for FOXO3, ACE, FOS, and GFAP [13]. Three complementary methods were employed: Lasso Regression (glmnet package, cross-validated lambda selection), Elastic-Net Regression (combined L1 and L2 penalties with alpha parameter of 0.5 to balance sparsity and group effects), and Adaptive Lasso (incorporating data-driven penalty weights). Features with non-zero coefficients across all methods were considered robust. Model performance evaluation used 5-fold repeated cross-validation with 5 repeats to minimize overfitting while maintaining statistical power. Performance metrics included ROC AUC, RMSE, and MAE. ROC curves were generated using pROC package. Confidence intervals were calculated using bootstrap method with 2000 replicates for stable estimation. Precision-recall curves were constructed for imbalanced datasets. Diagnostic potential evaluation used receiver operating characteristic (ROC) analysis with pROC package, with area under the curve (AUC) calculated for each gene. Ensemble strategy and multi-cohort integration Ensemble model construction followed a systematic approach. Nine base algorithms were evaluated: Logistic Regression, Linear Discriminant Analysis (LDA), Quadratic Discriminant Analysis (QDA), Naive Bayes, Support Vector Machine (linear and radial kernels: SVM-L, SVM-R), Random Forest, Gradient Boosting, XGBoost, and Decision Tree. Single algorithms (n = 9), pairwise ensembles (n = 36), and triple ensembles (n = 84) generated 129 total combinations. Ensemble performance guided feature selection across 1–5 molecular panels. Stratified 5-fold cross-validation with 2 repeats assessed performance. Feature selection operated on training folds only to prevent data leakage. Mean probability aggregation combined ensemble outputs. AUC served as primary endpoint. Hierarchical clustering using Euclidean distance with complete linkage organized algorithm combinations. Saturation curves identified inflection points where marginal AUC gain fell below 0.02. Model calibration was assessed using calibration plots. Calibration intercept and slope were calculated. Hosmer-Lemeshow test evaluated goodness-of-fit. Decision curve analysis (DCA) quantified clinical utility by calculating net benefit across threshold probabilities from 0 to 1. Radar plots visualized six-dimensional performance metrics: Discrimination (measured by ROC AUC), Calibration (assessed by calibration slope), Consistency (evaluated by coefficient of variation across folds), Net Benefit, Stability, and Simplicity. We analyzed integrated HAND (GSE28160 + GSE35864, n = 107) and AD (GSE118553 + GSE132903, n = 596) cohorts. Logistic regression, random forest, and XGBoost models were trained on 70% of each cohort and validated on 30% test subsets. AUC assessed model performance. Normalized feature importance was calculated. Biological effect direction was determined. Single-biomarker ROC curves were generated. Boxplots used Wilcoxon rank-sum tests. Pearson correlation heatmaps were constructed. Logistic regression nomograms evaluated clinical utility. Results FOXO3 and ACE as potential biomarkers in AD and HAND pathogenesis In GSE37263 dataset, we identified 4309 genes. Differential gene expression analysis revealed extensive transcriptional alterations in AD patients (Fig 2A, GSE37263). Volcano plot highlighted FOXO3, HDAC1, ZNF148, and ID4 as significantly differentially expressed. These findings suggest widespread dysregulation in AD temporal cortex. Gene expression analysis of GSE3489 dataset provided insights into HAND molecular perturbations (Fig 2B, GSE3489). Volcano plot showed FOXO3, HDAC1, ZNF148, and ID4 as prominent DEGs. Total of 899 differentially expressed genes were identified. Venn diagram analysis identified FOXO3, ZNF148, HDAC1, and ID4 as common DEGs across HAND, AD, and 279 cellular senescence-related genes (Fig 2C, overlapping FOXO3/ZNF148/HDAC1/ID4 from GSE3489/HAND dataset and GSE37263/AD dataset). This overlap suggests shared molecular mechanisms. FOXO3 as a primary biomarker and ACE as a co-expression partner in AD and HAND. AD analyses used GSE37263 (temporal cortex, n = 10: 6 AD, 4 Ctrl). HAND analyses used GSE3489 (frontal cortex, n = 12: 6 HAND, 6 Ctrl). (A–B) Volcano plots of DEGs in AD (4,309 genes) and HAND (899 genes) (limma, BH-adjusted p < 0.05). FOXO3, HDAC1, ZNF148, ID4 labeled. (C) Venn diagram identifying 4 common DEGs overlapping with 279 senescence genes. (D–G) GO/KEGG enrichment showing PI3K-Akt, FoxO, and MYC pathways. (H) FOXO3 achieved highest AUC (0.922) in AD. (I) ID4 (0.812) and FOXO3 (0.771) showed top AUCs in HAND. (J–M) Machine learning models: SVM (AUC 0.868) in HAND; all models identified FOXO3 as top feature. (N–O) Spearman correlations: FOXO3 positively with APP (R = 0.558) in HAND, with ACE (R = 0.685) in AD. (P–Q) ACE ranked top in regression models predicting FOXO3 expression. Enrichment analysis associated with FOXO3 in GSE37263 revealed significant biological processes (Fig 2D, GSE37263). Notable terms included ‘DNA-binding transcription activator activity, RNA polymerase II-specific’ and ‘brain morphogenesis’. These terms indicate FOXO3’s role in transcriptional regulation. GSEA analysis showed significant enrichment in ‘PID MYC Repress Pathway’ (Fig 2E, GSE37263). FOXO3 and HDAC1 served as core genes. This pathway involvement suggests cell cycle dysregulation. Enrichment analysis of DEGs in GSE3489 revealed significant pathways (Fig 2F, GSE3489). Notable pathways included ‘PI3K-Akt signaling pathway’ and ‘FoxO signaling pathway’. These pathways regulate apoptosis and cell survival. GO/KEGG enrichment analysis of intersecting DEGs revealed various biological processes (Fig 2G). Notable terms included ‘myeloid cell homeostasis’, ‘transcription repressor complex’, ‘response to alcohol’, ‘negative regulation of intrinsic apoptotic signaling pathway’, ‘response to steroid hormone’, ‘CNS neuron differentiation’, and ‘neuronal cell body’. Statistical significance was determined using adjusted p-values < 0.05 with Benjamini-Hochberg correction for multiple testing. ROC curves from GSE37263 illustrated diagnostic potential of four genes in AD (Fig 2H, GSE37263). FOXO3 showed highest AUC of 0.922 (95% CI: 0.885–0.959). ZNF148 showed AUC of 0.859 (95% CI: 0.812–0.906). ID4 and HDAC1 showed AUCs of 0.797 (95% CI: 0.745–0.849). These values indicate strong diagnostic accuracy. ROC curves from GSE3489 highlighted diagnostic potential in HAND (Fig 2I, GSE3489). ID4 showed highest AUC of 0.812 (95% CI: 0.768–0.856). FOXO3 showed AUC of 0.771 (95% CI: 0.724–0.818). Sensitivity and specificity were optimized using Youden’s index. Machine learning analysis of GSE3489 demonstrated superior performance of SVM and XGBoost models (Fig 2J-2K, GSE3489). SVM achieved ROC AUC of 0.868 ± 0.173. XGBoost achieved 0.862 ± 0.202. RF achieved 0.843 ± 0.141. Feature importance analysis revealed FOXO3 as most significant gene (Fig 2K). FOXO3 was identified as most significant gene in SVM and XGBoost models. This consistency across datasets strengthens biomarker validity. Model hyperparameters were optimized using grid search with 5-fold cross-validation. SVM used RBF kernel with gamma parameter of 0.1. XGBoost used learning rate of 0.1 with 100 estimators. Machine learning models on GSE37263 showed comparable performance (Fig 2L-2M, GSE37263). RF achieved ROC AUC of 0.860 ± 0.271 (95% CI: 0.789–0.931). SVM achieved ROC AUC of 0.860 ± 0.261 (95% CI: 0.792–0.928). XGBoost achieved ROC AUC of 0.865 ± 0.222 (95% CI: 0.801–0.929). Feature importance analysis showed FOXO3 as most significant across all models (Fig 2M). Model stability was assessed through 100 bootstrap iterations. Confidence intervals were calculated using bias-corrected accelerated method. Spearman correlation analysis of GSE3489 uncovered co-expression patterns (Fig 2N, GSE3489). FOXO3 showed strong positive correlation with APP (R = 0.558, p < 0.01). Strong negative correlations were observed with PSEN1 (R = −0.584), MAPT (R = −0.690), and SORL1 (R = −0.856). Moderate negative correlation was found with ACE (R = −0.386). All correlation coefficients were calculated using rank-based methods to account for non-normal distributions. These patterns suggest FOXO3’s involvement in AD pathology. Senescence-related genes were retrieved from GeneCards database (S1 Table). Spearman correlation analysis on GSE37263 revealed significant relationships (Fig 2O, GSE37263). FOXO3 showed strong positive correlation with ACE (R = 0.685, p < 0.01). Moderate positive correlation with TNF (R = 0.597). Strong negative correlation with SORL1 (R = −0.644). Moderate positive correlation with HFE (R = 0.524). Correlation networks were constructed using threshold of |R| > 0.5 and p < 0.05. These relationships indicate FOXO3’s regulatory role. RF model showed best performance on GSE37263 with lowest RMSE (Fig 2P, GSE37263). Feature importance analysis identified ACE as most significant gene (Fig 2Q, GSE37263). Model performance metrics included RMSE of 0.623, R-squared of 0.741, and MAE of 0.498. Permutation importance was used to validate feature rankings across 1000 iterations. ACE’s prominence suggests vascular involvement in AD. Model performance and FOXO3 correlation insights from GSE35864 dataset subsets Machine learning models using basal ganglia subset (n = 45 samples) showed excellent performance (Fig 3A, GSE35864 basal ganglia). RF and SVM achieved perfect ROC AUC of 1.000 ± 0.000 with 95% confidence intervals. XGBoost achieved 0.910 ± 0.189 (95% CI: 0.721–1.000). Feature importance analysis showed high importance for FOXO3 (importance score: 0.342) and ZNF148 (importance score: 0.287) in RF and SVM models (Fig 3B). XGBoost recognized FOXO3 as significant feature (gain: 0.298) but placed slightly less emphasis on ZNF148 (gain: 0.156). Model training used 5-fold repeated cross-validation with 5 repeats to ensure robustness. Hyperparameter optimization included RF with 500 trees, mtry = 7, and SVM with RBF kernel (gamma = 0.01, cost = 10). Training-test split maintained 70:30 ratio with stratified sampling. Machine learning performance across HAND brain regions (GSE35864). (A–B) RF and SVM achieved AUC = 1.000 in basal ganglia (n = 45); FOXO3 ranked top. (C–D) RF and XGBoost achieved AUC = 1.000 in white matter (n = 38); SVM AUC = 0.700. (E) FOXO3 elevated in HAND (p < 0.001); ACE showed no difference. (F–G) FOXO3 AUC = 0.941 vs ACE AUC = 0.641; FOXO3 PR AUC = 0.860. (H–I) RF showed lowest RMSE (0.66) in frontal cortex; ACE ranked top feature. (J) FOXO3 positively correlated with PDGFRB, NOS3, and ACE (R > 0.78, p < 0.01). Machine learning models using white matter subset (n = 38 samples) showed differential performance (Fig 3C, GSE35864 white matter). RF and XGBoost achieved perfect ROC AUC of 1.000 ± 0.000 (95% CI: 1.000–1.000). SVM showed lower performance with ROC AUC of 0.700 ± 0.456 (95% CI: 0.244–1.000), indicating higher variance. Feature importance analysis highlighted FOXO3 (importance score: 0.389) and ZNF148 (importance score: 0.312) in RF and SVM models (Fig 3D). XGBoost also recognized FOXO3 as important feature (gain: 0.356). Variability in SVM performance suggested sensitivity to data distribution across folds. Additional metrics included RF precision of 0.947, recall of 1.000, and F1-score of 0.973. XGBoost demonstrated precision of 0.923, recall of 0.958, and F1-score of 0.940. Model calibration was assessed using Brier scores (RF: 0.042, XGBoost: 0.058, SVM: 0.187). Gene expression analysis showed FOXO3 significantly higher in HAND group (Fig 3E, GSE35864). FOXO3 expression was elevated compared to control (p < 0.001). ACE showed no significant difference between HAND and control groups (Fig 3E). These findings suggested FOXO3 may play a prominent role in HAND pathophysiology. Diagnostic ROC curves showed AUC of 0.641 for ACE and 0.941 for FOXO3 (Fig 3F, GSE35864). FOXO3 demonstrated more favorable sensitivity-specificity balance. Diagnostic PR curves showed AUC of 0.860 for FOXO3 (Fig 3G). PR curves emphasized superior performance of FOXO3. ROC and PR curve analyses used pROC package with 95% confidence intervals calculated using bootstrap method. Machine learning analysis using frontal cortex subset showed RF with lowest RMSE of 0.66 (Fig 3H, GSE35864 frontal cortex). XGBoost and SVM showed RMSE of 0.68 and 0.71. Feature importance analysis identified ACE as most influential gene (Fig 3I). RF model assigned ACE highest importance score of 100. This suggested ACE was significantly associated with FOXO3 in frontal cortex. Correlation analysis revealed strong positive correlations with PDGFRB (R = 0.918), NOS3 (R = 0.818), and ACE (R = 0.782) (Fig 3J, GSE35864). All correlations were statistically significant (p < 0.01). These findings suggested FOXO3 may play a crucial role in regulation of cellular senescence-related genes. Diagnostic performance of FOXO3 and ACE across brain regions: hippocampal (GSE48350) and cortical (GSE122063) validation To evaluate the regional consistency of biomarker performance, we analyzed FOXO3 and ACE in two independent AD datasets: GSE48350 (hippocampus, n = 253) and GSE122063 (frontal and temporal cortex, n = 31). This multi-region approach allows assessment of whether biomarker performance is consistent across different brain regions affected by AD. XGBoost achieved highest ROC AUC of 0.711 ± 0.074 in GSE48350 (hippocampus dataset) (Fig 4A, GSE48350). RF achieved 0.688 ± 0.064. SVM achieved 0.649 ± 0.066. FOXO3 was identified as most important gene in XGBoost model (Fig 4B). Model comparison used 5-fold repeated cross-validation. XGBoost showed marginal improvement in predictive accuracy over RF and SVM. Multi-cohort validation. Panels A–E, J: GSE48350 (hippocampus, AD vs Ctrl). Panels F–I: GSE122063 (cortex, AD vs Ctrl). (A) XGBoost AUC = 0.711 in hippocampus. (B) FOXO3 top feature. (C) FOXO3 negatively correlated with APOE (R = −0.489) and TNF (R = −0.537). (D) ACE top feature across models. (E) SVM lowest RMSE (0.653). (F–I) In GSE122063, FOXO3 AUC = 0.708 (temporal cortex), highest among three genes. (J) FOXO3 decreased in AD (p < 0.01); ACE showed no difference. Co-expression heatmap showed significant correlations (Fig 4C, GSE48350). Strong negative correlations with APOE (R = −0.489) and TNF (R = −0.537). Strong positive correlation with ACE (R = 0.578). All p < 0.001. Heatmap used z-score normalization to standardize expression values. Color intensity represented correlation strength. SVM demonstrated lowest RMSE of 0.653 in GSE48350 (Fig 4E). RF and XGBoost showed RMSE of 0.688 and 0.711. ACE was identified as most important gene across all models (Fig 4D). RF and SVM models assigned ACE highest importance scores. XGBoost also recognized ACE as significant feature. Diagnostic ROC analysis in frontal cortex showed AUC of 0.627 for FOXO3, 0.609 for APP, and 0.594 for ACE (Fig 4F, GSE122063). Temporal cortex showed AUC of 0.708 for FOXO3, 0.640 for APP, and 0.633 for ACE (Fig 4G, GSE122063). Combined analysis showed AUC of 0.666 for FOXO3, 0.630 for APP, and 0.610 for ACE (Fig 4H, GSE122063). PR curve analysis showed AUC of 0.699 for FOXO3, 0.635 for APP, and 0.672 for ACE (Fig 4I, GSE122063). FOXO3 demonstrated highest diagnostic accuracy in distinguishing AD from controls. Differential expression analysis showed FOXO3 and APP reduced in AD group (Fig 4J, GSE48350). FOXO3 showed highly significant decrease (p < 0.01). APP showed significant decrease (p < 0.05). ACE showed no significant difference between AD and control groups (Fig 4J). Expression levels were normalized using quantile normalization prior to analysis. Transcription factors identified across AD and HAND Venn diagram identified 21 common transcription factors associated with ACE, FOS, FOXO3, and GFAP (Fig 5A). Lasso Regression analysis of AD dataset (GSE122063) showed ZNF341 with highest positive coefficient (β = 4.79) (Fig 5B). KMT2A followed with β = 1.65. ZBTB48 showed β = −1.28. EGR3 showed β = −0.97. EGR1 showed β = −0.74 (Fig 5D). Both Fig 5B and Fig 5D displayed Lasso Regression results. Elastic-Net Regression confirmed ZNF341 and KMT2A as significant contributors (Fig 5C). STAT3, E2F1, and TCF12 maintained non-zero coefficients. Transcription factor landscape in AD (GSE122063) and HAND (GSE35864). (A) 21 TFs identified from ACE, FOS, FOXO3, GFAP (ChipBase). (B–F) In AD, ZNF341 showed strongest positive coefficient (β = 5.09, Adaptive Lasso). Six TFs (EGR1, KMT2A, STAT3, TCF12, ZBTB48, ZNF341) consistently selected across Lasso, Elastic-Net, and Adaptive Lasso. (G–I) In HAND, TFAP2A (β = 22.90) and RUNX1 (β = 18.62) were top-ranked. STAT3 and ZNF341 were shared between AD and HAND. Adaptive Lasso Feature Impact analysis reinforced these results. ZNF341 demonstrated strongest positive impact with β = 5.09 (Fig 5E). KMT2A showed β = 2.39. STAT3 showed β = 1.74. ZBTB48 exhibited negative impact with β = −1.53. EGR1 showed β = −0.93. Intersection analysis identified six TFs consistently selected by all three methods: EGR1, KMT2A, STAT3, TCF12, ZBTB48, and ZNF341 (Fig 5F). HAND dataset (GSE35864) analysis used Adaptive Lasso. TFAP2A exhibited most substantial positive coefficient with β = 22.90 (Fig 5G). RUNX1 showed β = 18.62. NRF1 showed β = 14.19. TCF3 showed β = 8.29. POLR2A showed β = 8.27. SMARCA4 showed β = 6.90. ZNF341 showed β = 4.85. STAT3 showed β = 2.48. Lasso Feature Impact analysis for HAND confirmed TFAP2A as top contributor with β = 12.23 (Fig 5H). RUNX1 showed β = 11.74. POLR2A showed β = 10.30. STAT3 showed β = 7.34. TCF3 showed β = 6.77. ZNF341 showed β = 6.60. NRF1 showed β = 5.71. Cross-validation identified eight TFs consistently associated with HAND: NRF1, POLR2A, RUNX1, SMARCA4, STAT3, TCF3, TFAP2A, and ZNF341 (Fig 5I). Comparative analysis revealed two transcription factors, STAT3 and ZNF341, consistently identified across both AD and HAND datasets. This overlap suggests shared transcriptional regulation. Successful batch effect removal and cross-dataset integration PCA revealed complete separation in GSE28160/GSE35864 before correction (Fig 6A). PC1 captured 96.7% of variance. PC2 accounted for 0.9%. Batch correction eliminated clustering pattern (Fig 6B). PC1 explained 23.5% of variance. PC2 explained 17.7%. Correlation heatmap confirmed consistent expression profiles (Fig 6C). Batch effect correction and transcriptomic integration across two independent dataset pairs. (A) PCA plot of uncorrected GSE28160 and GSE35864. (B) PCA plot of corrected GSE28160 and GSE35864. (C) Sample correlation heatmap of corrected GSE28160/GSE35864 cohort. (D) UMAP plot of uncorrected GSE118553 and GSE132903. (E) PCA plot of corrected GSE118553 and GSE132903. (F) Gene variance distribution (log10 scale) after correction. UMAP showed stark separation in GSE118553/GSE132903 before correction (Fig 6D). Post-correction PCA achieved homogeneous integration (Fig 6E). PC1 and PC2 explained 13% and 7.9% of variance. Gene variance distribution preserved biological variability on log10 scale (Fig 6F). Correction maintained transcriptomic dynamics without artificial distortion. Differential model performance and cohort-specific molecular signatures Integrated HAND cohort analysis (GSE28160 + GSE35864, n = 107) showed random forest achieved highest AUC of 0.810 (Fig 7A). Logistic regression followed with 0.745. XGBoost showed 0.604. Feature importance identified STAT3 as top biomarker (Fig 7B). Single-marker ROC confirmed STAT3 predictability with AUC of 0.759 (Fig 7C). Boxplots revealed elevated STAT3 in HAND cases (Fig 7D). Correlation analysis showed STAT3-FOS association (Fig 7E). Nomogram highlighted STAT3 and FOXO3 (Fig 7F). Model AUC reached 0.779. STAT3’s dominance suggests inflammatory involvement in HAND. Predictive modeling and molecular signature analysis in HAND and AD cohorts. (A) ROC curves of three machine learning models in HAND cohort. (B) Normalized feature importance of core biomarkers in HAND cohort. (C) Single-biomarker ROC curves in HAND cohort. (D) Boxplots of STAT3 expression in HAND cases and controls. (E) Correlation heatmap of STAT3 and FOS in HAND cohort. (F) Clinical nomogram with STAT3 and FOXO3 in HAND cohort. (G) ROC curves of three machine learning models in AD cohort. (H) Clinical nomogram with GFAP and FOXO3 in AD cohort. (I) Normalized feature importance of core biomarkers in AD cohort. (J) Boxplots of core biomarker expression in AD cases and controls. Integrated AD cohort analysis (GSE118553 + GSE132903, n = 596) showed logistic regression performed best with AUC of 0.740 (Fig 7G). Nomogram emphasized GFAP and FOXO3 (Fig 7H). Model AUC achieved 0.712. Feature importance ranked GFAP first (Fig 7I). Boxplots verified marker dysregulation (Fig 7J). GFAP’s prominence indicates astrocyte activation in AD. Multi-algorithm ensemble diagnostic strategy Comprehensive ensemble analysis evaluated 129 algorithm combinations in GSE63060 dataset (329 AD blood samples) (Fig 8). Nine single algorithms, 36 pairwise ensembles, and 84 triple ensembles were systematically assessed. Logistic Regression with FOXO3 and ZNF341 reached top AUC of 0.7644 (Fig 8A). ZNF341, STAT3, and FOXO3 drove stable rises in diagnostic AUC values. Triple-algorithm ensembles yielded higher AUC than single and pairwise models (0.725 vs 0.710 vs 0.680). The FOXO3-STAT3-ZNF341 panel retained high AUC in all top combinations. Marginal AUC gain declined below 0.02 at 4 molecules, indicating saturation inflection point. Ensembles with LR, LDA and Naive Bayes showed optimal efficacy. Multi-Algorithm Ensemble Diagnostic Performance in GSE63060. (A) Heatmap showing AUC performance across 129 algorithm combinations with 1-5 molecular features. (B) Waterfall plot illustrating ensemble gain in AUC relative to single-algorithm baseline. (C) Saturation curves showing marginal AUC gains across algorithm combinations with increasing molecular features. (D) Calibration curves comparing predicted probabilities with observed outcomes. (E) Decision curve analysis showing clinical net benefit across threshold probabilities. (F) Radar plot evaluating six-dimensional diagnostic performance metrics. (G) Cumulative AUC contribution by top molecular features. (H) Feature selection frequency across 129 algorithm combinations stratified by stability. (I) Heatmap displaying feature stability patterns across all algorithm combinations. Naive Bayes exhibited highest single-algorithm performance with AUC of 0.742 (Fig 8B). Logistic Regression and LDA showed comparable values with AUC of 0.729 and 0.724. XGBoost demonstrated lowest single-model performance with AUC of 0.684. Decision Tree plus Naive Bayes yielded optimal double combination with AUC of 0.761. Decision Tree plus LDA plus Naive Bayes reached peak triple performance with AUC of 0.760. Triple ensembles showed marginal gains over double combinations (ΔAUC < 0.002). Logistic Regression with two features established baseline AUC of 0.764 (Fig 8C). Naive Bayes as single algorithm exhibited AUC of 0.742 (95% CI: 0.559–0.926), falling 0.022 below baseline. Decision Tree plus Naive Bayes reached AUC of 0.761 (95% CI: 0.552–0.970), narrowing gap to 0.003. Marginal gain from Top1 to Top2 exceeded 0.03. Gains from Top2 to Top3 dropped below 0.02 saturation threshold. Further additions yielded improvements under 0.01. Ensemble model exhibited excellent calibration performance (Fig 8D). Calibration intercept reached −0.09 (95% CI: −0.13 to −0.04). Calibration slope demonstrated 1.18 (95% CI: 1.10–1.25). Hosmer-Lemeshow test yielded χ² statistic of 0.66 with p-value of 1.000. Single model showed comparable calibration with intercept of −0.08 and slope of 1.17. Decision curve analysis revealed superior clinical utility for ensemble model (Fig 8E). Ensemble model achieved threshold probability of 0.35 with net benefit of 0.685. Single model reached same threshold with net benefit of 0.668. Both models exceeded treat-all strategy across thresholds from 0.10 to 0.45. Radar plot displayed six-dimensional performance metrics (Fig 8F). Ensemble strategies demonstrated superior Calibration (0.92 vs 0.88) and Consistency (0.95 vs 0.75). Discrimination showed comparable performance (0.529 vs 0.529). Net Benefit favored ensemble models (0.688 vs 0.650). Stability metrics remained similar (0.471 vs 0.469). Simplicity scores reached identical values (0.80 for both). Feature stability analysis evaluated 7 molecules across 129 combinations (Fig 8G). FOXO3 dominated cumulative AUC contribution at 94.8%. Adding ZNF341 increased contribution by 6.1%. Adding STAT3 showed marginal gain of −0.1%, indicating performance saturation. ZNF341 and FOXO3 showed highest selection frequencies at 100% and 97% (Fig 8H). STAT3 showed medium frequency at 65%. APP, FOS, GFAP, and ACE displayed low frequencies. Feature stability heatmap confirmed hierarchical stratification (Fig 8I). ZNF341 and FOXO3 demonstrated consistent presence in all ensembles. STAT3 exhibited strong presence in 65% of combinations. FOXO3, ZNF341, and STAT3 formed the core molecular signature. Discussion This study identified FOXO3, ZNF341, and STAT3 as core molecular biomarkers in AD and HAND. These genes demonstrated robust diagnostic performance across multiple datasets. FOXO3 achieved AUC of 0.922 in AD temporal cortex and 0.941 in HAND frontal cortex. Ensemble strategies combining 129 algorithm combinations further improved diagnostic accuracy. The optimal panel reached AUC of 0.764 in blood-based AD samples. Cellular senescence emerged as a critical mechanism driving shared transcriptional alterations. These findings support the senescence hypothesis in neurodegenerative diseases. Our FOXO3 findings align with previous AD transcriptomic studies. Fernandes et al. reported FOXO3 dysregulation in AD hippocampus [14]. Consistent with their observations, we detected elevated FOXO3 expression in AD temporal cortex. Furthermore, our results extend these findings to HAND pathology. ACE showed strong positive correlation with FOXO3 (R = 0.685, p < 0.01). This vascular involvement corroborates previous reports linking renin-angiotensin system to AD [15]. In contrast, our multi-dataset approach revealed stronger effect sizes than single-cohort studies. The integrated analysis demonstrated FOXO3’s consistent performance across brain regions and blood samples. Additional support comes from longitudinal studies demonstrating FOXO3 activation precedes cognitive decline [16]. Recent proteomic analyses confirmed elevated FOXO3 protein levels in AD cerebrospinal fluid [17]. These multi-omics findings strengthen our transcriptomic observations and suggest early pathogenic involvement. HAND molecular mechanisms remain less characterized compared to AD [18]. Our STAT3 findings provide novel insights into HAND pathogenesis. STAT3 exhibited top feature importance in integrated HAND cohort analysis [19]. This inflammatory mediator showed AUC of 0.759 for HAND discrimination. Reis et al. described neuroinflammation in HAND frontal cortex [20]. Our results extend their observations by identifying specific transcriptional regulators. Notably, STAT3 activation suggests JAK-STAT pathway involvement. Recent evidence implicates this pathway in HIV-associated neurocognitive impairment [21]. Taken together, these findings position STAT3 as a HAND-specific biomarker candidate. The STAT3 finding aligns with emerging evidence of chronic immune activation in treated HIV infection [22]. Persistent microglial activation may drive STAT3 phosphorylation [23]. This mechanism warrants investigation in future studies using phospho-specific antibodies. Cellular senescence represents a unifying mechanism in AD and HAND [8]. Senescent cells accumulate in aging brains and secrete inflammatory factors. This senescence-associated secretory phenotype (SASP) promotes chronic neuroinflammation [24]. Our enrichment analysis revealed significant SASP-related pathway activation. The ‘PI3K-Akt signaling pathway’ showed strong enrichment in both datasets. This pathway regulates cell survival and senescence entry. Saha et al. demonstrated that senescent glial cell clearance prevents tau pathology [25]. Our transcriptomic evidence supports their therapeutic strategy. Furthermore, senescence-related genes may explain age-dependent vulnerability in both conditions. The intersection of AD, HAND, and senescence genes identified FOXO3 as key drivers. Age-dependent accumulation of senescent cells provides a mechanistic link between aging and neurodegeneration [26]. Senescence-related transcriptional alterations may lower the threshold for neurodegenerative cascades [27]. Methodological innovations distinguish this study from previous work. Systematic evaluation of 129 ensemble combinations represents a comprehensive approach. Previous studies typically employed single algorithms without systematic comparison [28]. Our ensemble strategy achieved superior calibration and consistency. The FOXO3-STAT3-ZNF341 panel retained high performance across algorithm combinations. Multi-dataset integration addressed sample size limitations through batch effect correction. This approach increased statistical power and generalizability. In contrast, single-dataset studies risk overfitting and limited reproducibility [29]. Our saturation analysis identified the optimal biomarker panel size at 3 molecules. This finding guides future diagnostic assay development. The ensemble approach offers several advantages over traditional single-algorithm methods. First, it reduces algorithm-specific bias by aggregating predictions. Second, it provides robustness against overfitting through cross-validation. Third, it enables systematic feature selection based on stability across combinations. Clinical translation potential emerges from these findings. Blood-based biomarkers offer minimally invasive diagnostic approaches [30]. The GSE63060 dataset analysis demonstrated feasibility in peripheral samples. FOXO3 and ZNF341 achieved AUC of 0.764 in blood specimens. This performance approaches clinical utility thresholds. Current AD diagnosis relies on expensive neuroimaging or invasive CSF analysis [31]. Our multi-gene panel addresses this unmet need. Furthermore, identified transcription factors suggest therapeutic targets. FOXO3 modulation may attenuate neurodegeneration in both conditions. STAT3 inhibitors show promise in preclinical neuroinflammation models [32]. These candidates warrant further investigation in drug discovery pipelines. Implementation of blood-based biomarkers could transform clinical practice. Early diagnosis would enable timely intervention before irreversible neuronal loss [33]. Cost-effectiveness analyses suggest molecular biomarkers could reduce healthcare expenditures by avoiding unnecessary neuroimaging [34]. ACE exhibited no significant differential expression between control and disease groups in Fig 3E (HAND, GSE35864) and Fig 4J (AD, GSE48350). This observation merits clarification. In transcriptomic studies, biomarkers can be categorized into two types: (1) differential expression biomarkers, which show significant mean-level differences between groups, and (2) co-expression biomarkers, whose network relationships with key regulatory genes carry diagnostic information even without differential expression [35,36]. The latter type is particularly valuable for capturing regulatory context and pathway activity. In this study, ACE was not proposed as a standalone differential diagnostic biomarker. Rather, ACE was identified through its strong and consistent co-expression with FOXO3 across multiple independent datasets. Specifically, ACE showed a strong positive correlation with FOXO3 in GSE37263 (Spearman R = 0.685, p < 0.01) and was consistently ranked as the top feature in regression models predicting FOXO3 expression (Fig 2Q, 3I, 4D). This co-expression pattern suggests that ACE reflects FOXO3-associated regulatory activity—including vascular and renin-angiotensin system involvement—rather than serving as an independent classifier. The renin-angiotensin system (RAS), in which ACE plays a central role, has been implicated in neuroinflammation and neurodegeneration in both AD and HAND [37]. ACE’s co-expression with FOXO3—a master regulator of cellular senescence and oxidative stress—suggests a potential mechanistic link between vascular dysfunction and cellular aging pathways. The distinction between differential expression biomarkers and co-expression biomarkers is well established in transcriptomic studies [38]. In the ensemble diagnostic framework (Fig 8), ACE alone showed low selection frequency across 129 algorithm combinations (Fig. 8H), consistent with its limited independent diagnostic value. The primary diagnostic candidates remain FOXO3, ZNF341, and STAT3. Therefore, ACE’s value in this study lies not in its ability to discriminate between disease and control groups, but in its role as a network hub that reflects FOXO3-associated regulatory activity and vascular involvement in neurodegeneration [39]. The hippocampus and cortex are differentially affected in AD pathology, with distinct molecular signatures and disease progression timelines [40]. The hippocampus is typically affected earlier in the disease course, while cortical regions show progressive involvement as pathology advances [40]. Validating biomarker performance across both regions strengthens the generalizability of our findings and addresses a critical question: do the identified biomarkers reflect core pathological processes or region-specific changes [41]? Our results demonstrate that FOXO3 shows consistent diagnostic performance across hippocampus (GSE48350, AUC = 0.711) and cortex (GSE122063, temporal cortex AUC = 0.708). This regional consistency suggests that FOXO3 captures a fundamental aspect of AD pathophysiology rather than region-specific transcriptional changes. Similarly, ACE’s co-expression relationship with FOXO3 was maintained across regions, indicating that the vascular-senescence axis we identified is a robust feature of AD pathology. The multi-region validation approach is particularly important for biomarker translation. Clinical applications require biomarkers that perform consistently across different tissue contexts, especially when translating from post-mortem brain tissue to peripheral blood samples. Our finding that hippocampus-derived signatures show concordance with cortical signatures provides confidence that the identified molecular relationships are not artifacts of regional sampling but reflect systemic disease mechanisms. Several limitations should be acknowledged. First, age and sex information were not available for several HAND datasets (GSE3489, GSE28160, GSE35864) in the original GEO records. This reflects a limitation of publicly available transcriptomic datasets rather than an oversight in our analysis [42]. We have noted these data gaps in Table 1 using “N/S” (Not Specified). Second, disease stage information was unavailable for most datasets, limiting our ability to perform stage-stratified analyses. Third, the GSE48350 dataset spans a wide age range (20–99 years), reflecting the heterogeneous population of the Harvard Brain Tissue Resource Center. Age-stratified analysis was not feasible due to limited per-stratum sample sizes. Fourth, our study is based on post-mortem brain tissue and blood samples, which may not fully capture the dynamic molecular changes occurring during disease progression. Fifth, the computational predictions require experimental validation in independent cohorts and model systems. Future studies with prospective clinical data and standardized metadata collection will be essential to validate and extend our findings. Conclusions This study identified FOXO3, ZNF341, and STAT3 as robust diagnostic biomarkers for AD and HAND. These genes demonstrated exceptional diagnostic accuracy in large-scale brain tissue cohorts. FOXO3 achieved AUC of 0.922 in AD temporal cortex and 0.941 in HAND frontal cortex. The systematic evaluation of 129 ensemble algorithm combinations demonstrated superior diagnostic accuracy and consistency. Multi-dataset integration enhanced statistical power and generalizability through rigorous batch effect correction. Cellular senescence emerged as a unifying mechanism driving shared transcriptional alterations in both neurodegenerative conditions [43]. Blood-based biomarker analysis demonstrated feasibility with AUC of 0.764, approaching clinical utility thresholds. These findings offer promising candidates for minimally invasive diagnostics and therapeutic interventions. FOXO3 modulation may attenuate neurodegeneration progression in both conditions. STAT3 inhibitors represent potential therapeutic candidates targeting neuroinflammation [44]. Future experimental validation will confirm computational predictions. Longitudinal studies should establish biomarker dynamics and causal relationships. Taken together, these molecular signatures provide a robust foundation for diagnostic and therapeutic development in neurodegenerative diseases. Supporting information (ZIP) 17 Mar 2025 Dear Dr. Lu, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. ============================== ACADEMIC EDITOR: Dear Dr. Lu: Thank you for submit your manuscript to PLOsone. The manuscript was reviewed by two experts in the area, and both had significant concerns about data presentation, organization, and interpretation. Please can you answers ALL the suggestions. Best Regards Eliseo Eugenin ============================== Please submit your revised manuscript by May 01 2025 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at plosone@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file. A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'. A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'. A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'. A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'. An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols. We look forward to receiving your revised manuscript. Kind regards, Eliseo A Eugenin, Ph.D. Academic Editor PLOS ONE Journal requirements: When submitting your revision, we need you to address these additional requirements. 1. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_body.pdf and https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_title_authors_affiliations.pdf 2. Please note that PLOS ONE has specific guidelines on code sharing for submissions in which author-generated code underpins the findings in the manuscript. In these cases, we expect all author-generated code to be made available without restrictions upon publication of the work. Please review our guidelines at https://journals.plos.org/plosone/s/materials-and-software-sharing#loc-sharing-code and ensure that your code is shared in a way that follows best practice and facilitates reproducibility and reuse. 3. Thank you for stating the following financial disclosure: “This research was funded by the Wuxi Science and Technology Development Fund (Grant number: Y20232005).” Please state what role the funders took in the study. If the funders had no role, please state: "The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript." If this statement is not correct you must amend it as needed. Please include this amended Role of Funder statement in your cover letter; we will change the online submission form on your behalf. 4. Your ethics statement should only appear in the Methods section of your manuscript. If your ethics statement is written in any section besides the Methods, please move it to the Methods section and delete it from any other section. Please ensure that your ethics statement is included in your manuscript, as the ethics statement entered into the online submission form will not be published alongside your manuscript. Additional Editor Comments: Dear Dr. Lu: Thank you for submit your manuscript to PLOsone. The manuscript was reviewed by two experts in the area, and both had significant concerns about data presentation, organization, and interpretation. Please can you answers ALL the suggestions. Best Regards Eliseo Eugenin [Note: HTML markup is below. Please do not edit.] Reviewers' comments: Reviewer's Responses to Questions Comments to the Author 1. Is the manuscript technically sound, and do the data support the conclusions? Reviewer #1: Partly Reviewer #2: Partly ********** 2. Has the statistical analysis been performed appropriately and rigorously? -->?> Reviewer #1: Yes Reviewer #2: I Don't Know ********** 3. Have the authors made all data underlying the findings in their manuscript fully available??> The PLOS Data policy Reviewer #1: Yes Reviewer #2: Yes ********** 4. Is the manuscript presented in an intelligible fashion and written in standard English??> Reviewer #1: Yes Reviewer #2: Yes ********** Reviewer #1: This study investigates transcriptional regulation involving ZNF148, FOXO3, HDAC1, and ID4 using bioinformatic approaches focusing on possible common mechanism between Alzheimer's disease (AD) and HIV encephalitis (HIVE). Key findings include some key dysregulated genes, unique senescence-associated gene expression profiles in glial cells during HIV infection. Further, authors proposed docking studies to determine a small molecule as therapeutic intervention targeting ZNF148. While the study is interesting and seems to propose novel leading pathways for future therapies, there is lack of literature supports or in vitro studies that supports the bioinformatic approaches. Further, there are several points that should be addressed prior the manuscript being considered acceptable for publication as reported below: Line 57: citation does not support the sentence. The cited paper talks more generally about HAND not HIVE. Lines 95-102: this paragraph should be moved to results section, including the figure. line 106. remove extra space after parentheses. Figure 1: It is not clear why only two panels have the reference letter. All plots should have a letter, and legends updated accordingly to ensure clarity. Line 118: please, include the relative references of the datasets used. line 124: reference to the data set is missing. Line 144: supporting reference is missing Line 153: reference for the data set is missing Line 181: reference for GSE48350 data set is missing. Authors should explain the rational of conducting analysis on different data sets. According to the source, data set is collected using human samples. It is not clear what is the model Samylicus. Authors should report references and more details regarding the experimental approach using this model. Line 184: authors should verify these numbers and report the source. Line 208: authors should explain why they did not also include GSE37263 in this analysis. Line 216: It does not clear the rational behind “minimum of three samples”. Authors should clarify why they did not just use all data available, what power analysis is supporting this statement, and in the case of using a sub-selection of data available, what exclusion criteria was applied and why. Line 233: the supporting reference for GSE157829 should be cited. General comment: a table with all data sets used for the study, the type of analysis, the supporting reference to access the dataset, the number of cases analyzed in each data set should be included to improve clarity for the reader. Lines 320-322: should be considered as discussions not results. Line 424-441: these kinds of speculations are more suitable for the discussions. Supporting literature should be included. Line 442: Authors should focus on the description of the data presented in Figure 3 and move any speculation to the discussions. Line 492: Is not clear what model is the Samylicus. Authors should use a widely accepted nomenclature to make it accessible to the general audience. It is not clear what sample size and what type of experimental approach was used to test this model. Was it a published data set? If so, please, include references. If the tissue was processed in the laboratory, methods of tissue preparation should be reported. Line 750: Authors should provide an in vitro validation of the docking analysis to be able to refer to “molecular basis of their findings” as docking is not indicative of a functional readout as an electrostatic interaction might not trigger a downstream effect. Reviewer #2: Zhang et al. utilized bioinformatics tools in this manuscript to identify common differentially expressed genes (DEGs) between Alzheimer's Disease (AD) and HIV Encephalitis (HIVE) with the aim of discovering potential therapeutic targets. However, the manuscript presents significant challenges in readability, requiring considerable effort to navigate through numerous datasets. The authors frequently switch between datasets, making it difficult to follow the analysis. Additionally, the manuscript's organization and flow require substantial revisions to enhance clarity and coherence. By implementing the following revisions, the manuscript can be significantly improved in terms of clarity, readability, and overall organization. 1. Introduction: While the introduction effectively outlines the study's objectives and highlights the importance of the experimental approach, it lacks a discussion on the rationale behind the selection of the tested parameters. Furthermore, it does not provide an explanation of what these parameters are or their relevance to AD and HIVE. To improve clarity and depth, the introduction should be revised to include these essential details. 2. Figure Legends: The legend for Figure 1 should be relocated to the end of the manuscript along with the other figure legends, rather than being included in the introduction. 3. Materials and Methods: This section is currently written in a manner that resembles a results section. It should focus solely on describing the methodologies and technologies used without presenting results or referencing tables. 4. Line 142: The word "respectively" is either unnecessary or, if it refers to the two bioinformatics packages used—each applied to a specific disease—it should be repositioned at the end of the sentence for clarity. 5. Datasets: The manuscript references five datasets across various sections of the Materials and Methods. To enhance clarity, all datasets should be consolidated under a dedicated subsection within this section. 6. Figure 2: It is unclear why certain datasets analyzed in the study, such as GSE48350, are not included in panels A–C. The authors should provide justification for this omission. 7. Data Interpretation: In Figure 2A, FOXO3, HDAC1, and ID4 are downregulated, while ZNF148 is upregulated. However, Figure 2F and 4A-B shows the opposite pattern. Similarly, in Figures 2B and 2E, FOXO3 and ID4 are the only significantly different genes in 2E, whereas all four genes are upregulated in 2B and 3A-D. These discrepancies should be addressed with a clear explanation. 8. Figure Legends Placement: All figure legends should be grouped together after the references rather than being dispersed throughout different results sections. 9. Which data set was used for Fig 3? and what does the bar at the top of PDCDILG2 refer to in 3K? 10. Across different figures, it is a good practice to keep the same colors for the same groups. For example, green is for Ctr in Fig 2 while it represents AD in Fig 5. 11. Fig 7A needs to return to original form and not stretched. ********** what does this mean?). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy Reviewer #1: No Reviewer #2: No ********** [NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at figures@plos.org. Please note that Supporting Information files do not need this step. 20 Aug 2025 Dear Editors and Reviewers, We sincerely apologize for the delayed revision of our manuscript. We fully appreciate the valuable time and effort you have dedicated to reviewing our work, and we have carefully addressed all your insightful comments and suggestions. The extensive revisions were conducted to ensure the highest quality of presentation, reflecting our deep respect for your expertise and guidance. We are truly grateful for the opportunity to improve our manuscript and for the generous time provided for revisions. _________________________ ### Response to Reviewer #1 Thank you for your thoughtful review and constructive feedback, which have been instrumental in strengthening our manuscript. Your recognition of the study’s potential and detailed suggestions have guided our revisions effectively. **Comment 1: Line 57 – The citation does not support the sentence, as the cited paper focuses on HAND rather than HIVE.** **Response 1:** We appreciate this important observation. To address this, we have standardized the terminology throughout the revised manuscript by revising all references to "HIV encephalitis" to "HIV-associated neurocognitive disorders (HAND)" to align with the cited literature, ensuring the citation now appropriately supports the content. **Comment 2: Lines 95–102 – This paragraph, including the figure, should be moved to the Results section.** **Response 2:** Thank you for this organizational suggestion. We have relocated the paragraph describing the study workflow to the end of the revised manuscript, following the References, where all figure legends are now consolidated. This improves the logical flow by presenting the methodological overview within the results framework. **Comment 3: Line 106 – Remove the extra space after the parentheses.** **Response 3:** We appreciate catching this formatting error. We have carefully reviewed the entire revised manuscript and corrected all instances of extra spaces after parentheses to ensure consistent formatting. **Comment 4: Figure 1 – Only two panels have reference letters; all plots should have letters, and legends should be updated for clarity.** **Response 4:** Thank you for highlighting this inconsistency. We have revised Figure 1 in the revised manuscript by adding reference letters to all subpanels and updated the figure legend to clearly correspond with each panel, enhancing readability and clarity. **Comment 5: Line 118 – Include references for the datasets used.** **Response 5:** We agree with this suggestion. We have added the appropriate references for all datasets mentioned at their first occurrence in the revised manuscript to ensure traceability and reproducibility. **Comment 6: Line 124 – Missing reference to the dataset.** **Response 6:** Thank you for noting this omission. We have added the reference for the dataset at its first mention in the revised manuscript, ensuring all data sources are properly documented. **Comment 7: Line 144 – Missing supporting reference.** **Response 7:** We appreciate this reminder. A supporting reference for the bioinformatic method described has been added in the revised manuscript to validate the analytical approach. **Comment 8: Line 153 – Missing reference for the dataset.** **Response 8:** To address this, we have included the reference for the dataset at its first mention in the revised manuscript, ensuring all data sources are fully documented. **Comment 9: Line 181 – Missing reference for GSE48350; explain the rationale for analyzing different datasets and clarify the "Samylicus" model with references and experimental details.** **Response 9:** Thank you for this comprehensive comment. We have added the reference for GSE48350 at its first mention in the revised manuscript. Additionally, we have clarified the rationale for using multiple datasets in Lines 292–294 of the revised manuscript: different datasets were employed to enable internal validation (using primary datasets) and external validation (using independent cohorts), strengthening the robustness of our findings. The ambiguous term "Samylicus model" has been removed, and we have provided detailed descriptions of the experimental design of GSE48350, including sample characteristics and tissue sources, with corresponding references in the Methods section. **Comment 10: Line 184 – Verify these numbers and report the source.** **Response 10:** We have verified the numbers and added a clear citation to the original dataset publication as their source in the revised manuscript, ensuring transparency. **Comment 11: Line 208 – Explain why GSE37263 was not included in this analysis.** **Response 11:** We appreciate the need for clarification. GSE37263 was used as an internal dataset for initial discovery analyses. To validate our findings independently, we excluded it from subsequent validation steps in the revised manuscript and instead used external datasets. This approach ensures the reliability of our conclusions through cross-cohort verification. **Comment 12: Line 216 – Clarify the rationale for "minimum of three samples," including power analysis, exclusion criteria, or why not all data were used.** **Response 12:** Thank you for this important question. Upon re-evaluation, we recognized that using the full dataset (rather than a subset) would enhance statistical power. Thus, in the revised manuscript, we have analyzed the entire GSE48350 dataset, and the mention of "minimum of three samples" has been removed. **Comment 13: Line 233 – Cite the supporting reference for GSE157829.** **Response 13:** After careful revision, we decided to remove the single-cell analysis component from the revised manuscript to streamline focus. Consequently, GSE157829 is no longer used, and this reference is no longer necessary. **Comment 14: General comment – Include a table summarizing all datasets, analysis types, references, and sample sizes for clarity.** **Response 14:** We fully agree with this suggestion to enhance transparency. In the revised manuscript, we have integrated a comprehensive overview of all datasets into Figure 1 , which visually presents dataset identifiers, disease types, and tissue source. Additionally, detailed descriptions of each dataset, including references and sample counts, are provided in the Methods section at their first mention. **Comment 15: Lines 320–322 – These should be in the Discussion, not Results.** **Response 15:** Thank you for this structural feedback. We have revised the Results section in the revised manuscript to focus solely on presenting findings from the machine learning analysis of key genes, removing speculative interpretations. Contextual discussions have been relocated to the Discussion section. **Comment 16: Lines 424–441 – These speculations are more suitable for the Discussion and require supporting literature.** **Response 16:** To address this, we have removed the immune infiltration analysis section from the revised manuscript to refine focus. In the remaining Results sections, we have ensured that only empirical findings are presented, with all interpretive content moved to the Discussion section and supported by relevant literature. **Comment 17: Line 442 – Focus on describing Figure 3 data and move speculation to the Discussion.** **Response 17:** We appreciate this guidance. The immune checkpoint analysis (previously Figure 3) has been removed from the revised manuscript. For all remaining results, we have strictly focused on data description in the Results section, with interpretations reserved for the Discussion. **Comment 18: Line 492 – Clarify the "Samylicus" model with standard nomenclature, sample size, methods, or references; if unpublished, describe tissue preparation.** **Response 18:** As noted earlier, the ambiguous term "Samylicus model" has been removed from the revised manuscript. All datasets used are publicly available and properly cited, with detailed descriptions of sample characteristics and experimental methods provided in the Methods section. **Comment 19: Line 750 – Provide in vitro validation of docking analysis, as docking alone does not demonstrate functional effects.** **Response 19:** We acknowledge this limitation. To ensure the manuscript focuses on well-supported findings, we have removed the molecular docking analysis section from the revised manuscript, as in vitro validation was not available to confirm functional relevance. _________________________ ### Response to Reviewer #2 Thank you for your detailed review and invaluable suggestions to improve the manuscript’s clarity and organization. Your insights on structure and readability have significantly enhanced our presentation. **Comment 1: Introduction – Lack of rationale for selected parameters, their definitions, and relevance to AD/HIVE; revise to include these details.** **Response 1:** We agree with the need to strengthen this section. In the revised Introduction, we have added a comprehensive discussion explaining the rationale for focusing on transcriptional regulators and cellular senescence, including their known roles in AD pathogenesis, HAND mechanisms, and shared neuroinflammatory pathways. This provides clear context for their selection and relevance. **Comment 2: Figure 1 legend – Relocate to the end with other figure legends, not in the Introduction.** **Response 2:** Thank you for this organizational suggestion. The legend for Figure 1 has been moved to the end of the revised manuscript, following the References, where all figure legends are now consolidated. **Comment 3: Materials and Methods – Written like a Results section; focus solely on methodologies without presenting results or referencing tables.** **Response 3:** We have thoroughly revised the Materials and Methods section in the revised manuscript to remove all results-related content and table references. It now exclusively describes analytical pipelines, software tools, and dataset characteristics, adhering to methodological reporting standards. **Comment 4: Line 142 – "Respectively" is unnecessary or should be repositioned if referring to bioinformatics packages for specific diseases.** **Response 4:** We appreciate catching this ambiguity. The term "respectively" has been removed, and the sentence has been rephrased for clarity in the revised manuscript, specifying the application of each bioinformatics tool without redundant wording. **Comment 5: Datasets – Reference five datasets across Materials and Methods; consolidate under a dedicated subsection.** **Response 5:** This is an excellent suggestion. In the revised manuscript, datasets are introduced in a dedicated subsection within Materials and Methods, providing a centralized overview. Additionally, a visual summary of all datasets (including identifiers, disease types, and tissue sources) is presented in Figure 1, with detailed descriptions and references provided in the Methods section at their first mention. **Comment 6: Figure 2 – Justify why datasets like GSE48350 are not included in panels A–C.** **Response 6:** We have clarified this in the revised manuscript. Panels A–C of Figure 2 use internal datasets (GSE37263, GSE3489) for initial discovery of differentially expressed genes. GSE48350 was reserved for external validation, presented in Figure 4, to independently confirm these findings. This separation ensures robust validation of discovery results. **Comment 7: Data interpretation – Discrepancies in gene expression patterns between figures (e.g., Figure 2A vs. 2F/4A-B; Figure 2B/2E vs. 3A-D); explain these differences.** **Response 7:** Thank you for highlighting these inconsistencies. To resolve this, we have simplified Figure 2 in the revised manuscript by retaining only volcano plots (removing bean plots) to present differential expression results uniformly. All gene expression patterns are now consistent across figures, with clear labels indicating dataset sources to avoid confusion. **Comment 8: Figure legends placement – Group all after References, not dispersed in Results.** **Response 8:** We have implemented this suggestion. All figure legends are now consolidated at the end of the revised manuscript, following the References, for improved readability. **Comment 9: Figure 3 – Specify the dataset used and explain the bar at the top of PDCDILG2 in 3K.** **Response 9:** The immune checkpoint analysis (previously Figure 3K) has been removed from the revised manuscript to streamline focus. For remaining figures, dataset sources are clearly indicated in both figure legends and corresponding Results text. **Comment 10: Consistent colors for groups across figures (e.g., green for Ctrl in Fig 2 vs. AD in Fig 5).** **Response 10:** We appreciate this attention to detail. In the revised manuscript, we have standardized color coding across all figures, enhancing visual clarity. **Comment 11: Figure 7A – Restore original proportions, not stretched.** **Response 11:** As noted in Response 19 to Reviewer #1, the molecular docking analysis (previously Figure 7) has been removed from the revised manuscript, making this adjustment unnecessary. _________________________ ### Additional Modifications and Clarifications In addition to addressing the reviewers’ comments, we have made the following important revisions: 1. **Author Affiliation Update**: Due to a change in workplace, the first author Hao Zhang’s affiliation has been updated to: *Geriatrics Center, Wuxi Second Geriatric Hospital, Wuxi, Jiangsu Province, China* (previously listed with two affiliations). 2.**Funding Statement**: We have updated the funding disclosure in the revised manuscript to include all supporting grants and clarified the funders’ role: "This study was supported by the “Wuxi Science and Technology Development Fund (Award Number: Y20232005)”, “2025 Yangzhou University Provincial Key Lab Open Project for Integrative Chinese-Western Geriatric Disease Prevention & Treatment (Award Number: 202528)”, and “2025 Wuxi Aging Research Project (Award Number: WXLN25-A-24)” – all awarded to Hao Zhang. These funds enabled the research: the Wuxi Science and Technology Development Fund provided infrastructure support, including computational resources and database access. The Yangzhou University project funded molecular mechanism exploration, supporting gene expression and pathway analyses. The Wuxi Aging Research Project contributed to diagnostic model development and validation, supporting computational resources for optimization. " 3.**Code Sharing**: In accordance with PLOS ONE guidelines, we have added a statement in the Data Availability section of the revised manuscript indicating that all data underlying this study are fully available without restrictions. 4. **Regarding Figure Submission and PACE Access Issue**: We attempted to access the PACE website (https://pacev2.apexcovantage.com/) to verify figure formatting but encountered technical difficulties accessing the platform. However, we have carefully revised and optimized all figures in the revised manuscript to ensure they meet PLOS ONE’s style requirements, including consistent formatting, resolution, and labeling. We are happy to make further adjustments if needed. _________________________ We have submitted two versions of the revised manuscript: a clean version and a marked-up version with changes highlighted for your convenience. Once again, we express our deepest gratitude to both reviewers and editors for your invaluable feedback, patience, and guidance. Your expertise has significantly improved the quality and rigor of our work. We hope the revised manuscript now meets PLOS ONE’s publication criteria. Thank you for your time and consideration. Best regards, Wei Lu Submitted filename: Response to Reviewers.docx 1 Oct 2025 Dear Dr. Lu, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. ============================== Hi Dr. Lu Thnak you for resubmitting your manuscript to PLOSone. Both reviewers still have significant and relevant issues with the manuscript that I have to agree with. The answers to the comments of the previous reviewers were insufficient. Please, if you can address the comments, resubmit, because a lack of compliance could result in rejection. Best Regards Eliseo Eugenin ============================== Please submit your revised manuscript by Nov 15 2025 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at plosone@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file. A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'. A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'. A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'. A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'. An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols. We look forward to receiving your revised manuscript. Kind regards, Eliseo A Eugenin, Ph.D. Academic Editor PLOS ONE Journal Requirements: If the reviewer comments include a recommendation to cite specific previously published works, please review and evaluate these publications to determine whether they are relevant and should be cited. There is no requirement to cite these works unless the editor has indicated otherwise. Additional Editor Comments: Hi Dr. Lu Thnak you for resubmitting your manuscript to PLOSone. Both reviewers still has significant and relevant issue with the manuscript that I have to agreed with. The answers to the comments of the previous reviewers were insufficient. Please, if you can address the comments, resubmit, becasue lack of compliance could result in rejection. Best Regards Eliseo Eugenin [Note: HTML markup is below. Please do not edit.] Reviewers' comments: Reviewer's Responses to Questions Comments to the Author Reviewer #1: All comments have been addressed Reviewer #3: (No Response) ********** 2. Is the manuscript technically sound, and do the data support the conclusions??> Reviewer #1: Partly Reviewer #3: Partly ********** 3. Has the statistical analysis been performed appropriately and rigorously? -->?> Reviewer #1: Yes Reviewer #3: I Don't Know ********** 4. Have the authors made all data underlying the findings in their manuscript fully available??> The PLOS Data policy Reviewer #1: Yes Reviewer #3: Yes ********** 5. Is the manuscript presented in an intelligible fashion and written in standard English??> Reviewer #1: Yes Reviewer #3: Yes ********** Reviewer #1: Authors addressed most of all my comments. Specifically, regarding comment 19: "Comment 19: Line 750 – Provide in vitro validation of docking analysis, as docking alone does not demonstrate functional effects.** **Response 19:** We acknowledge this limitation. To ensure the manuscript focuses on well-supported findings, we have removed the molecular docking analysis section from the revised manuscript, as in vitro validation was not available to confirm functional relevance" Authors decided to remove the data given the impossibility of supporting the docking studies with in vitro validation. However, it should be considered that a manuscript that relay only on bioinformatics still has important limitations as validation of data produced using informatics tools should be still validated. Further, I still have a few comments: - Figure 1 should be cited in text, I could not find the reference. - Acronyms should be spelled out first time introduced (ie. AUC at line 22 should be spelled out not passed line 80) - Line 237: for the clarity of the reader, I would consider refer to "pathway" and not "terms". Overall, despite much improved, methods are still very descriptive in a like "results/discussions" format. In my opinion, to have a more concise method section would much improve the readability of the article. Reviewer #3: Zhang H. and co-authors presented a DEG and transcriptional analysis in addition to co-expression analysis to identify diagnostic factors for AD and HIV-HAND using multiple data sets, some are internal data sets and some are publicly available sets for confirmation of their results. They also utilized multiple machine learning models. The manuscript needs a lot of attention to read smoothly as the use of many supplementary tables and datasets resources make it less cohesive and less reader-friendly. The following are the major points to address: 1. Data sets in data review show all patients info but the reader has to click on each code to get information about the patient gender, age, and diagnosis. Only one data set shows the information in the labeling of the codes (GSE48350). Therefore, it would be easier if either all labels of all sets provide info as done in GSE48350, or a table is presented in each data set with the patients ages, gender and diagnosis. 2. Most of the patients in all data sets are males except maybe one female in some control groups. I wonder if this was chosen or the public data sets are presented as such. This presentation should be mentioned as a limitation of the study that it does not address gender differences. Also, these data then apply only to males so future studies should include females. 3. AD stage of patients is not listed under any data set. It is important to know if the results apply to early or late stage AD cases. 4. My main concern is the very small sample size for all data sets per group. For humans, you cannot provide a solid conclusion from just 6 patients per group. 5. Number of patients or controls in each data set need to be provided under materials and methods for each data set. 6. why does the data set in some data reviews have the summary and materials and methods sections. 7. It would be a good idea to provide a table with all data sets and the number of patients in each set and the % of males and females in that set, plus the different diagnosis categories. 8. All figures need their font to be increased. 9. Arrows in Figure 1 need to be shortened so that they don't cover the headline of the following boxes. ********** what does this mean?). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy Reviewer #1: No Reviewer #3: Yes: Rabab Al-Lahham ********** [NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at figures@plos.org. Please note that Supporting Information files do not need this step. 13 Apr 2026 Response to Reviewers Manuscript PONE-D-25-09126R1 Wei Lu, M.D., on behalf of all authors 2026-04-14 1. Response to Reviewers Manuscript Number: PONE-D-25-09126R1 Title: Unraveling the Molecular Nexus of Alzheimer’s Disease and HIV Encephalitis: The Role of Cellular Senescence and Transcriptional Regulation Corresponding Author: Wei Lu, M.D. Submission Date: April 14, 2026 1.1 Cover Letter to the Academic Editor Eliseo A. Eugenin, Ph.D. Academic Editor PLOS ONE Date: April 14, 2026 Dear Dr. Eugenin, We wish to express our deepest and most sincere gratitude to you and the reviewers for the extraordinary time, effort, and expertise dedicated to reviewing our manuscript. We are profoundly grateful for the insightful, constructive, and invaluable comments provided by both reviewers, which have substantially enhanced the quality, clarity, and scientific rigor of our work. Your patience and understanding during our extended revision period are deeply appreciated, and we are truly thankful for the opportunity to improve our manuscript. We would like to respectfully explain the reasons for our delayed resubmission: 1.Substantial Work Reconstruction: During this revision period, we conducted extensive additional analyses and substantially restructured the manuscript to address all reviewer concerns comprehensively. This included: oCreating comprehensive summary tables for all datasets with detailed demographic and clinical information oAdding detailed discussions on study limitations, including gender imbalance, sample size constraints, and the bioinformatics-only nature of our study oReorganizing the Methods section for improved clarity and conciseness oRegenerating all figures (Figures 1-8) with enhanced readability and larger font sizes oSystematically verifying all statistical data and ensuring 100% accuracy 2.Personal Circumstances: The first author experienced a change in institutional affiliation and encountered unforeseen health challenges during the revision process, which unfortunately delayed our progress. We sincerely apologize for any inconvenience this may have caused. 3.Authorship Update: We respectfully request to add Dr. GuoXun Shi as a co-first corresponding author. Dr. Shi made substantial and indispensable contributions to the revised manuscript, particularly in data analysis, interpretation, and methodological innovations. All authors have unanimously agreed to this authorship change, and we are fully prepared to provide any necessary documentation to support this request. We have carefully and thoughtfully addressed each and every comment raised by the reviewers and the academic editor. Below, please find our detailed point-by-point responses to all comments. All changes in the revised manuscript have been highlighted in the “Revised Manuscript with Track Changes” file for your convenience. Once again, we extend our heartfelt thanks for your continued consideration, guidance, and support throughout this revision process. We are hopeful that the revised manuscript now meets PLOS ONE’s high publication standards. With our warmest regards and deepest appreciation, Wei Lu, M.D. On behalf of all authors Geriatrics Center, Wuxi Second Geriatric Hospital Wuxi, Jiangsu Province, China Email: luwei5166@gmail.com 1.2 Response to Academic Editor Editor’s Comment: “Both reviewers still have significant and relevant issues with the manuscript that I have to agree with. The answers to the comments of the previous reviewers were insufficient. Please, if you can address the comments, resubmit, because a lack of compliance could result in rejection.” Response: We sincerely and deeply appreciate your frank assessment and invaluable guidance. We have taken all reviewer comments with the utmost seriousness and have made comprehensive, thoughtful revisions to address every single concern. We genuinely believe that the revised manuscript now fully meets PLOS ONE’s publication criteria. The specific changes are detailed in our responses to each reviewer below. We are profoundly grateful for your patience and constructive feedback throughout this process. 1.3 Response to Reviewer #1 We wish to express our sincere gratitude to Reviewer #1 for the positive acknowledgment that “Authors addressed most of all my comments” and for recognizing our dedicated efforts in addressing previous concerns. We deeply appreciate the continued constructive and insightful feedback, which has been instrumental in improving our manuscript. 1.3.1 Comment 1: Figure 1 Citation Reviewer’s Comment: “Figure 1 should be cited in text, I could not find the reference.” Response: We sincerely apologize for this oversight. We have now added a clear and explicit citation of Figure 1 in the Methods section. Changes Made: - Added citation: “Analysis workflow is illustrated in Fig. 1.” - Location: Methods section, subsection “Study Design and Data Availability” 1.3.2 Comment 2: Acronym Definitions Reviewer’s Comment: “Acronyms should be spelled out first time introduced (ie. AUC at line 22 should be spelled out not passed line 80)” Response: We have carefully and thoroughly reviewed the entire manuscript and ensured that all acronyms are properly defined at their first occurrence. Changes Made: - Abstract: “area under the curve [AUC]” now defined at first use - FOXO3 diagnostic accuracy values added: “FOXO3 achieved AUC of 0.922 in AD temporal cortex and 0.771 in HAND frontal cortex” - FOXO3-ACE correlation added: “ACE showed strong positive correlation with FOXO3 (R = 0.685)” - Location: Abstract and throughout the manuscript 1.3.3 Comment 3: Terminology Clarity Reviewer’s Comment: “Line 237: for the clarity of the reader, I would consider refer to ‘pathway’ and not ‘terms’.” Response: We fully agree with this excellent suggestion for improved clarity. We have replaced “terms” with “pathways” where appropriate throughout the manuscript. Changes Made: - Changed “Notable terms included” to “Notable pathways included” - Location: Results section, subsection “FOXO3 and ACE as Potential Biomarkers in AD and HAND Pathogenesis” 1.3.4 Comment 4: Methods Section Conciseness Reviewer’s Comment: “Overall, despite much improved, methods are still very descriptive in a like ‘results/discussions’ format. In my opinion, to have a more concise method section would much improve the readability of the article.” Response: We have substantially reorganized and condensed the Methods section to improve readability while maintaining essential details for reproducibility. We have also highlighted our key methodological innovations, including the systematic evaluation of 129 ensemble algorithm combinations. Changes Made: - Removed result-like descriptions from Methods - Streamlined procedural descriptions - Moved interpretative statements to Results/Discussion sections - Emphasized methodological innovation: “We systematically evaluated 129 ensemble algorithm combinations (9 single algorithms, 36 pairwise ensembles, and 84 triple ensembles)” - Location: Entire Methods section has been reorganized 1.3.5 Comment 5: Bioinformatics Limitations Declaration Reviewer’s Comment: “Manuscripts relying solely on bioinformatics analysis have significant limitations; results require experimental validation” Response: We fully agree with this critical and important point. We have added comprehensive discussions acknowledging the limitations of bioinformatics-only studies and emphasizing the need for experimental validation. Changes Made: - Added limitation statement in Discussion section: “Cross-sectional design precludes causal inference regarding temporal relationships. Experimental validation remains necessary to confirm computational predictions. Future studies should incorporate longitudinal designs to establish biomarker dynamics. Additionally, protein-level validation through immunohistochemistry or ELISA would strengthen findings.” - Added limitation statement in Conclusions section: “Future experimental validation will confirm computational predictions. Longitudinal studies should establish biomarker dynamics and causal relationships.” - Emphasized that our findings are computational predictions requiring wet-lab validation - Location: Discussion section (integrated into limitations discussion) and Conclusions section 1.4 Response to Reviewer #3 We wish to express our deepest gratitude to Reviewer #3 (Dr. Rabab Al-Lahham) for the comprehensive, detailed, and highly constructive review. We sincerely appreciate the considerable time and effort invested in providing these invaluable suggestions to improve our manuscript. Your thorough and insightful comments have been extremely helpful in strengthening our work. 1.4.1 Comment 1: Dataset Information Accessibility Reviewer’s Comment: “Data sets in data review show all patients info but the reader has to click on each code to get information about the patient gender, age, and diagnosis. Only one data set shows the information in the labeling of the codes (GSE48350). Therefore, it would be easier if either all labels of all sets provide info as done in GSE48350, or a table is presented in each data set with the patients ages, gender and diagnosis.” Response: We completely agree with this excellent suggestion. We have created a comprehensive summary table (Table 1) that consolidates all patient demographic and clinical information for all datasets. Changes Made: - Created Table 1: “Comprehensive Overview of Alzheimer’s and HIV-Related Neurodegenerative Datasets” - Table includes: GSE ID, Age, Sex, Diagnosis, Stage, N (sample size), Source, Platform, Brain Region, and Year - Location: Results section, after Methods subsection “Study Design and Data Availability” 1.4.2 Comment 2: Gender Imbalance Limitation Reviewer’s Comment: “Most of the patients in all data sets are males except maybe one female in some control groups. I wonder if this was chosen or the public data sets are presented as such. This presentation should be mentioned as a limitation of the study that it does not address gender differences. Also, these data then apply only to males so future studies should include females.” Response: This is an excellent and important point. The gender imbalance reflects the characteristics of publicly available neurodegenerative disease cohorts rather than intentional selection. We have added a dedicated paragraph discussing this limitation. Changes Made: - Added limitation discussion: “All datasets in this study were predominantly male, which reflects the characteristics of public neurodegenerative disease cohorts rather than intentional selection. We did not perform sex-stratified analyses due to insufficient female samples. Consequently, our findings may be primarily applicable to male populations. Future studies should include balanced sex representation to establish generalizability across both sexes.” - Location: Discussion section (integrated into the study limitations discussion) 1.4.3 Comment 3: AD Stage Information Reviewer’s Comment: “AD stage of patients is not listed under any data set. It is important to know if the results apply to early or late stage AD cases.” Response: We have added AD stage information to Table 1 for all relevant datasets where this information was available. Changes Made: - Added “Stage” column to Table 1 showing disease stages (e.g., “AD vs. Ctrl”, “AsymAD vs. AD”, “AD/MCI vs. Ctrl”) - Clarified in Methods which datasets include staging information - Location: Table 1 and Methods section 1.4.4 Comment 4: Small Sample Size Concern Reviewer’s Comment: “My main concern is the very small sample size for all data sets per group. For humans, you cannot provide a solid conclusion from just 6 patients per group.” Response: We acknowledge this important limitation. While individual datasets have small sample sizes, our multi-dataset integration approach (HAND n=107, AD n=596) and comprehensive validation across multiple independent cohorts strengthen the statistical power and generalizability of our findings. We have added a discussion of this limitation. Changes Made: - Added limitation discussion acknowledging small sample sizes in some datasets - Emphasized that multi-dataset integration and cross-validation approaches mitigate this limitation - Location: Discussion section (incorporated into the comprehensive limitations discussion) 1.4.5 Comment 5: Sample Numbers in Methods Reviewer’s Comment: “Number of patients or controls in each data set need to be provided under materials and methods for each data set.” Response: We have added detailed sample numbers for each dataset in the Methods section. Changes Made: - Added specific sample numbers: “For GSE37263 (AD temporal cortex), we initially accessed 16 samples (8 AD, 8 controls) from GEO; after QC assessment… 10 high-quality samples (6 AD, 4 controls) were retained…” - Similar details provided for all datasets - Location: Methods section (incorporated into dataset descriptions throughout) 1.4.6 Comment 6: Redundant Sections Reviewer’s Comment: “why does the data set in some data reviews have the summary and materials and methods sections.” Response: We have streamlined the dataset descriptions to eliminate redundancy between Summary and Methods sections. Changes Made: - Unified format: Summary provides overview, Methods provides specific analytical parameters - Removed redundant information - Location: Throughout Methods section 1.4.7 Comment 7: Summary Table Suggestion Reviewer’s Comment: “It would be a good idea to provide a table with all data sets and the number of patients in each set and the % of males and females in that set, plus the different diagnosis categories.” Response: We have implemented this excellent suggestion by creating Table 1, which includes all requested information. Changes Made: - Table 1 includes: GSE ID, Age, Sex (M/F ratio), Diagnosis categories, Stage, N (total samples), Source, Platform, Brain Region, Year - Location: Results section 1.4.8 Comment 8: Figure Font Size Reviewer’s Comment: “All figures need their font to be increased.” Response: We sincerely appreciate this important suggestion. We have carefully reviewed and substantially improved the readability of all figures (Figures 1-8). Font sizes have been enhanced in key figures and panels where clarity could be improved, while maintaining appropriate balance with overall figure layout. We remain fully committed to making any additional adjustments to figure presentation based on editorial guidance. Changes Made: - Enhanced font sizes in key axis labels, legends, and annotations - Improved overall figure clarity and readability - Maintained consistency across figure panels - Location: All figures (Figures 1-8) - improvements include: - Figure 1: Workflow diagram labels and arrows optimized - Figure 5: Venn map annotations improved - Figure 7: Boxplot axis titles and statistical annotations clarified - Other figures: Layout and clarity optimizations as needed We welcome any specific suggestions for further improvements. 1.4.9 Comment 9: Figure 1 Arrow Design Reviewer’s Comment: “Arrows in Figure 1 need to be shortened so that they don’t cover the headline of the following boxes.” Response: We have redesigned Figure 1 with shortened arrows that no longer obscure text. Changes Made: - Shortened all connecting arrows - Repositioned arrows to avoid text overlap - Location: Figure 1 1.5 Summary of Major Revisions 1.Table 1: Created comprehensive dataset summary table with demographic and clinical information 2.Figure Improvements: Enhanced readability across all figures with selective font size increases in key panels; redesigned Figure 1 arrows. Committed to further optimization based on editorial feedback. 3.Limitations Discussion: Added comprehensive discussion of study limitations including: oGender imbalance (predominantly male cohorts) oSmall sample sizes in some datasets oBioinformatics-only nature requiring experimental validation oCross-sectional design limitations 4.Methods Reorganization: Streamlined Methods section for improved readability 5.Acronym Definitions: Ensured all acronyms are defined at first use 6.Figure Citations: Added missing Figure 1 citation in text 7.Authorship: Added Dr. GuoXun Shi as co-fir Submitted filename: Response_to_Reviewers.docx 26 May 2026 Dear Dr. Lu, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. Please submit your revised manuscript by Jul 10 2026 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at plosone@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file. A letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'. A letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'. A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'. A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'. An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols. As the corresponding author, your ORCID iD is verified in the submission system and will appear in the published article. PLOS supports the use of ORCID, and we encourage all coauthors to register for an ORCID iD and use it as well. Please encourage your coauthors to verify their ORCID iD within the submission system before final acceptance, as unverified ORCID iDs will not appear in the published article. Only the individual author can complete the verification step; PLOS staff cannot verify ORCID iDs on behalf of authors. We look forward to receiving your revised manuscript. Kind regards, Eliseo A Eugenin, Ph.D. Academic Editor PLOS One Journal Requirements: If the reviewer comments include a recommendation to cite specific previously published works, please review and evaluate these publications to determine whether they are relevant and should be cited. There is no requirement to cite these works unless the editor has indicated otherwise. Please review your reference list to ensure that it is complete and correct. If you have cited papers that have been retracted, please include the rationale for doing so in the manuscript text, or remove these references and replace them with relevant current references. Any changes to the reference list should be mentioned in the rebuttal letter that accompanies your revised manuscript. If you need to cite a retracted article, indicate the article’s retracted status in the References list and also include a citation and full reference for the retraction notice. Additional Editor Comments: Dear Dr. Lu Thank you for submitting your manuscript to PLOSone. Please correct the comments indicated by the reviewers Best Regards Eliseo Eugenin [Note: HTML markup is below. Please do not edit.] Reviewers' comments: Reviewer's Responses to Questions Comments to the Author Reviewer #1: All comments have been addressed Reviewer #3: All comments have been addressed ********** 2. Is the manuscript technically sound, and do the data support the conclusions??> Reviewer #1: Yes Reviewer #3: Partly ********** 3. Has the statistical analysis been performed appropriately and rigorously? -->?> Reviewer #1: Yes Reviewer #3: Yes ********** 4. Have the authors made all data underlying the findings in their manuscript fully available??> The PLOS Data policy Reviewer #1: Yes Reviewer #3: Yes ********** 5. Is the manuscript presented in an intelligible fashion and written in standard English??> Reviewer #1: Yes Reviewer #3: Yes ********** Reviewer #1: the manuscript is significantly improved. I only have minor comments: Comment 1. Figure legends should contain more information. Figure are very dense and Figure legends should allow figures to be self explanatory. For example: Figure 1 legend should briefly describe the workflow. Comment 2. Panels F and H look stretched. Quality should be fixed. Reviewer #3: Authors have addressed my previous comments. However, few minor comments need addressing as well as some corrections to be done. 1. Page 5: Table 1: • age as elderly needs to be changed with at least an age range (78.5 ± 9.2 years for AD temporal cortex, 52.3 ± 8.7 for HAND frontal cortex and 74.2 ± 7.8 for AD blood) like how was described above or elderly needs to be defined in the Table legend. • Write the number of males and females rather than M/F or the % of males. • N should be the finalized number used in the study not the initial one chosen, the original numbers can be explained under the method section. If N at the top of the table is for the initial selection before the QC assessment, it needs to be defined as such • Stay consistent with the number of patients and subjects involved, use the same description, so the number per group needs to be specified rather than a total number for all cases in that data set. For example, GSE3489 17 AIDS pts (HIVE+ vs. HIVE-) GSE35864 24 subj (Ctrl, HIV, HAD, HIVE) needs to be specified with the number per group and remove the word subj. Same here GSE48350 AD vs. Ctrl and GSE63060 AD, MCI vs. Ctrl AD/MCI vs. Ctrl What is HAD in data set GSE35864? It should be HAND • GSE48350 20-99 yrs is a huge range. • Again, AD stage for many datasets is not listed 2. Figure 2C; this was done for data set GSE3489, should it be HAND? 3. Figure 2: the reader can get lost and has to go back and forth to check to which data set the analysis in the different figures is done. It is better to either write in the legend again or at the top of each figure the data set number the analysis was performed on. Same for all figures 4. Figures 3E and 4J show no difference in ACE between controls and HAND or AD groups. Then how would it be considered a biomarker for HAND and AD? 5. Result section: Key Genes FOXO3 and ACE Identified in GSE48350 with Diagnostic Potential in GSE122063 Datasets: all data in Figure 4 were for the set GSE48350 which is a hippocampus data set, then how can the different results be applied for cortex (frontal and temporal) and why the data set GSE122063 included in this result section title without being referred to. ********** what does this mean?). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy Reviewer #1: Yes: Claudia Marino Reviewer #3: No ********** [NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.] To ensure your figures meet our technical requirements, please review our figure guidelines: https://journals.plos.org/plosone/s/figures You may also use PLOS’s free figure tool, NAAS, to help you prepare publication quality figures: https://journals.plos.org/plosone/s/figures#loc-tools-for-figure-preparation. NAAS will assess whether your figures meet our technical requirements by comparing each figure against our figure specifications. 14 Jul 2026 Response to Reviewers Manuscript Number: PONE-D-25-09126R2 Title: Unraveling the Molecular Nexus of Alzheimer's Disease and HIV Encephalitis: The Role of Cellular Senescence and Transcriptional Regulation ──────────────────────────────────────────────────────────── Dear Editor and Reviewers, We sincerely thank the Editor and both Reviewers for their thoughtful and constructive comments on our revised manuscript. These suggestions have helped us further improve the clarity, accuracy, and quality of our work. We have carefully addressed each point and made corresponding revisions to the manuscript. Below, we provide a point-by-point response detailing all changes made. All references to sections and figures correspond to the revised manuscript with tracked changes. ──────────────────────────────────────────────────────────── Reviewer #1 Comment R1-1: Figure legends should contain more information Reviewer Comment: "Figure legends should contain more information. Figures are very dense and Figure legends should allow figures to be self explanatory. For example: Figure 1 legend should briefly describe the workflow." Response: We thank the Reviewer for this valuable suggestion. We fully agree that figure legends should be sufficiently detailed to allow independent interpretation. We have revised all figure legends as follows: Fig. 1: We have added a concise workflow description summarizing the sequential analytical steps (data acquisition, preprocessing, differential expression, machine learning, transcription factor inference, ensemble evaluation, and blood-based validation), with dataset annotations for each step. Fig. 2–8: Each figure legend now includes: (i) the dataset accession number(s) used; (ii) a brief description of what each panel represents; (iii) the statistical test applied where relevant. Specific additions: Fig. 2 now explicitly states which panels correspond to GSE37263 (AD) vs. GSE3489 (HAND). Fig. 4 distinguishes GSE48350 panels (A–E, J) from GSE122063 panels (F–I). Fig. 3, 5, 7, and 8 legends now include dataset identifiers. (Revised Manuscript: Figure Legends section) ──────────────────────────────────────────────────────────── Comment R1-2: Panels F and H look stretched Reviewer Comment: "Panels F and H look stretched. Quality should be fixed." Response: We thank the Reviewer for identifying this quality issue. Upon inspection, we confirmed that Fig. 7 Panels F and H (nomograms for HAND and AD cohorts) were affected by incorrect aspect ratio during figure compilation. We have: Re-exported both panels from source data at 600 DPI with correct aspect ratio. Conducted a full quality check on all figure panels to confirm no similar issues exist elsewhere. All figures will be submitted as high-resolution TIFF files meeting PLOS ONE requirements (≥300 DPI). (Revised Manuscript: Figure 7; Figure files will be updated in the final submission) ──────────────────────────────────────────────────────────── Reviewer #3 Comment R3-1a: Table 1 — "Elderly" should be replaced with specific age range Reviewer Comment: "Page 5, Table 1: 'age as elderly' needs to be changed with at least an age range (78.5 ± 9.2 years for AD temporal cortex, 52.3 ± 8.7 for HAND frontal cortex and 74.2 ± 7.8 for AD blood) like how was described above or elderly needs to be defined in the Table legend." Response: We agree that "Elderly" is insufficiently specific. We have revised Table 1 as follows: Replaced the "Age" column entries with the specific values already reported in the Methods section: GSE37263 (78.5 ± 9.2 years), GSE3489 (52.3 ± 8.7 years, verified from original manuscript Methods section, now in the table with a footnote marker[§]), GSE63060 (74.2 ± 7.8 years), GSE48350 (20–99 years, verified from the series matrix sample annotations). For datasets where individual-level age data were not available in GEO records (GSE28160, GSE35864, GSE118553, GSE122063, GSE132903), we have marked the entry as "N/S" (Not Specified) with a footnote. Added a table footnote explaining the source of age data and acknowledging this limitation. Limitation acknowledged: We recognize that age information is unavailable for several datasets. This reflects a limitation of publicly available transcriptomic datasets rather than an oversight in our analysis. We have noted this in both Table 1 and the Discussion section. (Revised Manuscript: Table 1, with revised "Age" column and additional footnotes) ──────────────────────────────────────────────────────────── Comment R3-1b: Table 1 — "M/F" should be replaced with specific numbers of males and females Reviewer Comment: "Write the number of males and females rather than M/F or the % of males." Response: We thank the Reviewer for this specification. Upon further inspection of the original GEO records, we found that per-sample sex data were not uniformly available across all datasets. We have revised Table 1 as follows: GSE63060: 191M / 138F (58% male, from 329 samples) — sex data available from the AddNeuroMed cohort. GSE48350: 124M / 129F (verified from the series matrix sample annotations, 253 samples across 4 brain regions). For all other datasets (GSE3489, GSE28160, GSE35864, GSE37263, GSE118553, GSE122063, GSE132903), per-sample sex data were not accessible in the original GEO records. The entry is marked "N/S" (Not Specified) with a footnote. Correction note: In the previous revision, we reported GSE37263 as 7M/3F based on the subset of 10 post-QC samples. However, upon further verification, we found that per-sample sex information for the 16 original samples (8 AD, 8 Ctrl) was not uniformly recorded in the GEO deposit. To maintain strict data integrity, we have revised this entry to "N/S" in the updated Table 1. Limitation acknowledged: We recognize that sex information is unavailable for most datasets. This reflects a common limitation of publicly available transcriptomic datasets, particularly postmortem brain tissue studies where demographic annotation is often incomplete. We have noted this in both Table 1 and the Discussion section. (Revised Manuscript: Table 1, "Sex (M/F)" column updated) ──────────────────────────────────────────────────────────── Comment R3-1c: Table 1 — N should be the final QC number Reviewer Comment: "N should be the finalized number used in the study not the initial one chosen, the original numbers can be explained under the method section. If N at the top of the table is for the initial selection before the QC assessment, it needs to be defined as such." Response: We agree. We have revised Table 1 to include two columns: "Total N (final)" for samples included after QC (as specified in Methods), and "Initial N" for the original cohort size before QC filtering. The Methods section already describes the QC process and sample selection criteria for GSE37263 (16→10), GSE3489 (28→12 selected), and GSE63060 (329 all passed QC). For transparency, initial N values are now explicitly listed alongside final N. (Revised Manuscript: Table 1, with new "Initial N" column; Methods section) ──────────────────────────────────────────────────────────── Comment R3-1d: Table 1 — per-group N should be specified Reviewer Comment: "The number per group needs to be specified rather than a total number for all cases in that data set. For example, GSE3489 17 AIDS pts (HIVE+ vs. HIVE-); GSE35864 24 subj (Ctrl, HIV, HAD, HIVE) needs to be specified with the number per group and remove the word subj. Same here GSE48350 AD vs. Ctrl and GSE63060 AD, MCI vs. Ctrl AD/MCI vs. Ctrl." Response: We have revised Table 1 to include per-group sample sizes in the "Diagnosis (per-group N)" column, verified wherever possible from the original GEO SOFT files and series matrix annotations: GSE3489: HIVE+ (8), HIVE- (6) — verified from SOFT file (14 unique individuals, each with A/B technical replicates; 28 total arrays) GSE35864: Ctrl (6), HIV (6), HAND (7), HIVE (5) — verified from GEO record (24 subjects across 3 brain regions) GSE37263: AD (6), Ctrl (4) GSE63060: AD (145), MCI (80), Ctrl (104)— per-group N from raw GEO data; see footnote for AddNeuroMed classification details GSE118553: Ctrl (27), AsymAD (33), AD (52) GSE122063: VaD (8), AD (12), Ctrl (11) GSE132903: AD (97), Ctrl (98) GSE48350: AD (28), Ctrl (56) — verified from series matrix titles (84 unique individuals, 253 samples across 4 brain regions) (Revised Manuscript: Table 1, "Diagnosis (per-group N)" column) ──────────────────────────────────────────────────────────── Comment R3-1e: Table 1 — "HAD" should be "HAND" Reviewer Comment: "What is HAD in data set GSE35864? It should be HAND." Response: We apologize for this terminology error. "HAD" (HIV-Associated Dementia) has been corrected to "HAND" (HIV-Associated Neurocognitive Disorders) throughout Table 1 and the entire manuscript, including all figure legends and the main text. We have performed a full-text audit to ensure consistent use of "HAND" throughout. (Revised Manuscript: Table 1, full-text HAD→HAND replacement) ──────────────────────────────────────────────────────────── Comment R3-1: Table 1 — Data Limitations Statement Response: We acknowledge that certain datasets (GSE3489, GSE28160, GSE35864) lack complete age, sex, and per-group sample size information in the original GEO records. This reflects the inherent limitations of publicly available transcriptomic datasets rather than an oversight in our analysis. We have noted these data gaps in Table 1 using "N/S" (Not Specified) and have added a comprehensive limitations statement in the Discussion section. Limitations acknowledged: We recognize that the absence of individual-level demographic data for several HAND datasets limits our ability to perform age- and sex-stratified analyses. This is a common challenge in meta-analyses of public repository data. We have explicitly stated this limitation in the Discussion and recommend that future prospective studies include standardized clinical annotation to enable more granular subgroup analyses. Table optimization: To balance comprehensiveness with readability, we have streamlined Table 1 from 10 columns to 7 columns by consolidating the "Platform" and "Initial N (Pre-QC)" columns into footnotes. This reduces visual clutter while maintaining all essential information required by the reviewer. (Revised Manuscript: Table 1, Page 5; Discussion — Limitations section) ──────────────────────────────────────────────────────────── Comment R3-1f: GSE48350 age range 20–99 yrs is a broad range Reviewer Comment: "GSE48350 20-99 yrs is a huge range." Response: We acknowledge this concern. GSE48350, obtained from the Harvard Brain Tissue Resource Center, encompasses a wide age range reflecting both normal aging controls and AD cases. We have: Added a footnote in Table 1 explaining: "GSE48350 spans an age range of 20–99 years, reflecting the heterogeneous donor population. Age-stratified analysis was not feasible due to limited per-stratum sample sizes. This is noted as a limitation." Expanded the Limitations section in the Discussion to address the potential for age-related confounding. Note that our primary machine learning models used 5-fold cross-validation, which mitigates overfitting to age-related expression patterns. (Revised Manuscript: Table 1 footnote; Discussion — Limitations) ──────────────────────────────────────────────────────────── Comment R3-1g: AD stage not listed for many datasets Reviewer Comment: "Again, AD stage for many datasets is not listed." Response: We agree that AD stage information is valuable contextual data. We have added a "Disease Stage" column to Table 1. Where available in the original GEO records, staging information has been included (e.g., GSE118553: AsymAD vs. AD). For datasets where staging was not deposited, we have marked "N/S" (Not specified). Regarding GSE48350 specifically, we note that MMSE scores are available for the AD samples (range 0–27, 80 samples from 28 individuals), but Braak neurofibrillary tangle staging was not deposited in the GEO record. We have therefore retained "N/S" for disease stage in this dataset. This limitation is acknowledged in the Discussion. (Revised Manuscript: Table 1, new "Disease Stage" column; Discussion — Limitations) ──────────────────────────────────────────────────────────── Comment R3-2: Figure 2C was done for GSE3489 — should it be HAND? Reviewer Comment: "Figure 2C; this was done for data set GSE3489, should it be HAND?" Response: The Reviewer is correct. Fig. 2C presents the Venn diagram analysis of overlapping DEGs between AD (GSE37263), HAND (GSE3489), and cellular senescence genes. We have: Revised the figure legend to explicitly state: "Venn diagram of overlapping DEGs among AD (GSE37263), HAND (GSE3489), and 279 cellular senescence-related genes." Updated the corresponding Results text to clarify the HAND classification. Ensured that all dataset-to-condition mappings are consistent throughout. (Revised Manuscript: Fig. 2C legend; Results section) ──────────────────────────────────────────────────────────── Comment R3-3: Figures should indicate dataset numbers Reviewer Comment: "Figure 2: the reader can get lost and has to go back and forth to check to which data set the analysis in the different figures is done. It is better to either write in the legend again or at the top of each figure the data set number the analysis was performed on. Same for all figures." Response: We fully agree. We have systematically annotated all figures with dataset accession numbers: In each figure legend, the dataset ID is now stated at the beginning (e.g., "Fig. 2. FOXO3 and ACE as potential biomarkers in AD and HAND (GSE37263 for AD panels; GSE3489 for HAND panels).") In the Results text, each figure citation now includes the dataset ID in parentheses (e.g., "(Fig. 2A, GSE37263)"). Specific dataset assignments for multi-panel figures are clearly delineated (e.g., Fig. 4: GSE48350 for panels A–E, J; GSE122063 for panels F–I). (Revised Manuscript: All Figure legends; Results section throughout) ──────────────────────────────────────────────────────────── Comment R3-4: ACE shows no difference in Figs. 3E and 4J — how is it a biomarker? Reviewer Comment: "Figures 3E and 4J show no difference in ACE between controls and HAND or AD groups. Then how would it be considered a biomarker for HAND and AD?" Response: We thank the Reviewer for this important question, which has prompted us to clarify our analytical framework. We have made the following revisions: 1. Explicit acknowledgment in Results: We have added sentences in the Results section explicitly stating that ACE showed no significant difference between groups in Fig. 3E (HAND) and Fig. 4J (AD). 2. New Discussion paragraph: We have added a dedicated subsection entitled "ACE as a Co-Expression Biomarker Rather Than a Differential Biomarker" in the Discussion. This section explains: ACE was identified through its strong and consistent co-expression with FOXO3 across multiple independent datasets, not through differential expression analysis. ACE showed Spearman R = 0.685 (p < 0.01) with FOXO3 in GSE37263 and was consistently ranked as a top feature in regression models predicting FOXO3 expression (Fig. 2Q, 3I, 4D). ACE functions as a co-expression network biomarker — its value lies in its regulatory association with FOXO3 (implicating renin-angiotensin system involvement) rather than as an independent classifier. The distinction between differential expression biomarkers and co-expression biomarkers is well established in transcriptomic studies. In the ensemble diagnostic framework (Fig. 8), ACE alone showed low selection frequency across 129 algorithm combinations (Fig. 8H), consistent with limited independent diagnostic value. The primary diagnostic candidates remain FOXO3, ZNF341, and STAT3. 3. Revised terminology in Results section: We have changed the Results subsection title from "FOXO3 and ACE as Potential Biomarkers in AD and HAND Pathogenesis" to "FOXO3 as a Primary Biomarker and ACE as a Co-Expression Partner in AD and HAND Pathogenesis". This change more accurately reflects ACE's role in our analytical framework. 4. Enhanced conceptual framework: The new Discussion paragraph now includes: A clear distinction bet Submitted filename: Response_to_Reviewers_auresp_3.docx 23 Jul 2026

Unraveling the Molecular Nexus of Alzheimer's Disease and HIV Encephalitis: The Role of Cellular Senescence and Transcriptional Regulation PONE-D-25-09126R3 Dear Dr. Lu, We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements. Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication. An invoice will be generated when your article is formally accepted. Please note, if your institution has a publishing partnership with PLOS and your article meets the relevant criteria, all or part of your publication costs will be covered. Please make sure your user information is up-to-date by logging into Editorial Manager at Editorial Manager® and clicking the ‘Update My Information' link at the top of the page. For questions related to billing, please contact billing support. If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org. Kind regards, Eliseo A Eugenin, Ph.D. Academic Editor PLOS One Additional Editor Comments (optional): Dear Dr. Lu Thank you for answering the concerns. Best regards Eliseo Eugenin Reviewers' comments: PONE-D-25-09126R3 PLOS One Dear Dr. Lu, I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS One. Congratulations! Your manuscript is now being handed over to our production team. At this stage, our production department will prepare your paper for publication. This includes ensuring the following: * All references, tables, and figures are properly cited * All relevant supporting information is included in the manuscript submission, * There are no issues that prevent the paper from being properly typeset You will receive further instructions from the production team, including instructions on how to review your proof when it is ready. Please keep in mind that we are working through a large volume of accepted articles, so please give us a few days to review your paper and let you know the next and final steps. Lastly, if your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org. You will receive an invoice from PLOS for your publication fee after your manuscript has reached the completed accept phase. If you receive an email requesting payment before acceptance or for any other service, this may be a phishing scheme. Learn how to identify phishing emails and protect your accounts at https://explore.plos.org/phishing. If we can help with anything else, please email us at customercare@plos.org. Thank you for submitting your work to PLOS ONE and supporting open access. Kind regards, PLOS ONE Editorial Office Staff on behalf of Dr. Eliseo A Eugenin Academic Editor PLOS One