External transportability of bacterial- and viral-associated host-response modules across public transcriptomic cohorts Authors: Reuben S. Maghembe, Guichuan Lai, Benjamin Liu, Benjamin Liu Published: 2026 ABSTRACT ObjectivesTo determine whether bacterial- and viral-associated whole-blood host-response modules discovered in GSE211567 retain their prespecified directions when applied unchanged to two external cohorts, including a second cross-platform cohort, and to quantify their robustness to analytical choices and gene deletion.MethodsFive biologically guided modules were defined and fixed in GSE211567 before external analysis. The same modules were applied unchanged to GSE73461 and GSE72810 using an unweighted mean gene-wise z-score rule. Primary contrasts included 52 bacterial and 94 viral samples in GSE73461 and 23 definite bacterial and 28 definite viral samples in GSE72810. Wilcoxon tests were adjusted across the five modules using the Benjamini-Hochberg method. Hodges-Lehmann shifts, rank-biserial effects and 10,000-replicate bootstrap confidence intervals were calculated. Sensitivity analyses assessed z-score reference populations, probable-case inclusion, probe summarisation, GSVA scoring and exhaustive leave-one/two-gene deletion.ResultsAll five modules retained their expected directions in both projection cohorts. BACT_M2, VIR_M1a, VIR_M1b and VIR_M2 had confidence intervals excluding zero and passed false-discovery-rate correction in both cohorts, whereas BACT_M1 remained directionally concordant but borderline in GSE73461 and GSE72810 (adjusted P = 0.0799 in each). All 30 z-reference, case-definition and probe-collapse sensitivity estimates retained the expected direction. Under GSVA, BACT_M1 gained statistical support, whereas VIR_M2 retained its viral-higher direction but lost confidence-interval and adjusted-P-value support. Across 29,826 leave-one/two-gene variants, every variant retained the expected direction and the minimum Pearson correlation with its complete-module score was 0.9940.ConclusionsThe predefined host-response modules retained their expected directions across two external cohorts measured on different Illumina platforms, with the strongest reproducible support for BACT_M2 and the three viral-associated modules. BACT_M1 and the GSVA behaviour of VIR_M2 demonstrate that direction preservation does not imply uniform statistical support across cohorts or scoring algorithms. Introduction Distinguishing bacterial from viral infection remains important for infectious-disease medicine, antimicrobial stewardship and host-response biology [1–4]. Although pathogen detection is central to diagnosis, microbiological results may be delayed, insensitive or difficult to interpret when colonisation, co-infection or previous antimicrobial exposure complicate pathogen attribution [2,5,6]. Host transcriptomic studies have therefore been widely explored as complementary approaches, but much work has emphasised diagnostic classifiers and prediction performance rather than whether the underlying host-response biology is transportable across heterogeneous cohorts [7–10]. A complementary question is whether biologically interpretable response programmes discovered in one setting behave similarly when tested in a different cohort. In this study, transportability means preservation of the prespecified biological direction when the same module definition and scoring rule are applied to another dataset. Fixed-module projection means applying predefined gene sets to an external dataset without changing their genes, expected directions or weights and without retraining a model. This is distinct from diagnostic validation. A classifier may perform well because of cohort-specific, technical or sampling structure, whereas the present analysis asks whether the same biological modules retain their expected bacterial-higher or viral-higher behaviour outside the discovery dataset [8,10,11]. Public infection transcriptomic cohorts are valuable for this purpose, but samples may differ by geography, clinical syndrome, pathogen spectrum, platform, preprocessing and case definition [7,9,11]. Examining the discovery contrast separately within the available geographic strata can reduce the risk that a pooled bacterial-versus-viral signal is driven mainly by one setting. The study therefore kept discovery and external testing separate: module definitions, expected directions and scoring rules were fixed before either external cohort was analysed. Here, GSE211567 was used to discover bacterial- and viral-associated host-response programmes while requiring cross-site directional support. Five conservative modules were defined before external testing and then applied unchanged to GSE73461 and GSE72810. The latter provided a second accession-level, sample-level and cross-platform cohort. The aim was not to train or validate a diagnostic classifier, but to test whether predefined modules retained their expected directions, quantify cross-cohort effect sizes and determine sensitivity to score reference, probe handling, case definition, alternative gene-set scoring and gene deletion. Methods Study design and datasets This study used a staged design that kept discovery separate from external testing. GSE211567 was used for discovery, biological interpretation and definition of the five modules. GSE73461 was analysed only after the module genes, expected directions and scoring rules had been fixed. GSE72810 was subsequently analysed as a second accession-level and sample-level cohort providing cross-platform validation. Neither external dataset was used to reselect genes, rename modules, alter module composition, tune weights or train a diagnostic classifier. Public host-transcriptomic datasets were considered if they contained infection-relevant human transcriptomic data, recoverable sample metadata, interpretable pathogen-class labels, usable feature identifiers and sufficient sample structure for the intended analysis. GSE161731 was retained only as a technical rehearsal dataset, while GSE261482 and GSE68310 were audited but not selected for formal projection. GSE211567, GSE73461 and GSE72810 were obtained from the Gene Expression Omnibus [12–14]. GSE211567 discovery and module definition GSE211567 provided the discovery dataset and contained whole-blood transcriptomic samples from Sri Lanka and the United States [12]. After metadata and expression-quality auditing, the prespecified primary discovery set contained 224 samples: 101 bacterial and 123 viral. The primary limma model evaluated bacterial versus viral infection while adjusting for site and sequencing batch, and separate bacterial-versus-viral models were also fitted within Sri Lanka and the United States. For each modelled feature, directional concordance was defined as agreement in the sign of the bacterial-versus-viral log2 fold change between the contrasts being compared. Percentage directional concordance was the number of same-sign features divided by the number of compared features, multiplied by 100. Spearman correlations of log2 fold changes were used as a complementary measure of ranked effect-size agreement across the pooled and site-specific analyses. Differential-expression results were treated as a discovery-ranking layer rather than a diagnostic signature [15]. Positive log2 fold-change values were interpreted as bacterial-higher and negative values as viral-higher. Transcript-level features were mapped to gene identifiers, summarised at gene level and carried into Gene Ontology biological-process enrichment [16,17]. Cross-site directionally supported features were prioritised, overlapping or redundant enriched terms were grouped, and candidate biological programmes were reviewed using the documented curation hierarchy. Module selection was therefore a biologically guided and reproducible curation step rather than mathematical optimisation for separation or prediction accuracy. The external-cohort results were not used to choose or modify module genes. Five modules were fixed before external testing: BACT_M1, BACT_M2, VIR_M1a, VIR_M1b and VIR_M2. GSE73461 formal external projection GSE73461 expression, annotation, metadata and group labels were audited independently of module discovery [13]. The cohort was measured on the Illumina GPL10558 platform. The 52 DefiniteBacterial, 94 DefiniteViral and 55 Control samples (n = 201) constituted the main z-score reference population. The primary inferential contrast remained restricted to the 52 DefiniteBacterial versus 94 DefiniteViral samples; Control samples contributed to the reference population but not to the bacterial-versus-viral test. Inflammatory, Kawasaki and Unknown groups were excluded from both the main z-score reference population and the primary contrast. Illumina probes were mapped to gene symbols, and locked genes were checked against predefined module-coverage thresholds. The numbers of scored genes were 24/25 for BACT_M1, 21/21 for BACT_M2, 128/128 for VIR_M1a, 33/33 for VIR_M1b and 105/106 for VIR_M2. GSE72810 cross-platform validation and probe selection GSE72810 contained 146 paediatric whole-blood samples measured using the Illumina HumanHT-12 v3 platform [14]. All 146 samples were retained in the main z-score reference population, whereas the prespecified primary inferential contrast was restricted to 23 definite bacterial versus 28 definite viral samples. Seventeen probable bacterial and seven probable viral samples were reserved for expanded-case sensitivity analysis. Sixteen controls and 55 uncertain samples contributed to the main z-score reference population but were excluded from the primary bacterial-versus-viral test. Predefined module genes were reconciled to the GSE72810 platform through Entrez identifiers. Of 313 module-gene instances, 303 were mapped and 10 were unmapped. Module coverage was BACT_M1 24/25, BACT_M2 20/21, VIR_M1a 125/128, VIR_M1b 33/33 and VIR_M2 101/106. When multiple authorised probes represented the same Entrez gene, a prespecified rule selected the probe with the highest median expression across all 146 samples, with lexicographic ordering used only to resolve exact ties. Probe selection was completed before testing group differences. Module scoring and statistical analysis For the primary mean-z analyses, expression for each mapped gene g in sample i was standardised within the prespecified reference population as z_gi = (x_gi – mean_g)/ SD_g. For a module containing K mapped and available genes, the sample-level module score was the unweighted arithmetic mean, score_i = (1/K) sum_g z_gi. Thus, every available gene contributed equally to the primary module score. Missing genes were omitted only after module coverage had been documented. No module was retrained, reweighted or direction-flipped. Wilcoxon rank-sum tests compared bacterial and viral module scores, with Benjamini-Hochberg correction across the five modules [18]. Positive effects denote bacterial-higher scores and negative effects denote viral-higher scores. In addition to median bacterial-minus-viral differences, the revision analyses calculated Hodges-Lehmann location shifts and rank-biserial effects. Confidence intervals were obtained using 10,000 bootstrap replicates. Sensitivity and robustness analyses Z-reference sensitivity was evaluated by repeating scoring using only the primary bacterial and viral samples as the reference population. GSE72810 sensitivity analyses additionally included probable bacterial and viral cases and compared the locked representative-probe scores with mean scores across all authorised probes. Score concordance was assessed using Pearson and Spearman correlations. A GSVA sensitivity analysis was performed in GSE73461 using the unchanged module gene sets and both z-reference populations. Because mean-z and GSVA scores have different numerical scales, cross-method interpretation focused on effect direction, rank-biserial effect, confidence intervals and adjusted P values rather than direct comparison of raw score magnitudes. Gene-deletion robustness was assessed exhaustively in GSE73461 by recalculating every module after removal of each individual gene and every pair of genes. Variant scores were compared with the corresponding complete-module scores, and expected-direction retention, Wilcoxon results and Pearson and Spearman correlations were recorded. Cross-cohort interpretation and reproducibility boundaries The GSE72810 and GSE73461 GEO sample accession sets were disjoint and the cohorts were measured on different Illumina array platforms. However, direct participant overlap could not be assessed because participant identifiers were not deposited, and the studies arose from the same broad investigator network. GSE72810 is therefore described as a second accession-level and sample-level cross-platform cohort rather than as a fully investigator-independent replication cohort. Analysis scripts, decision logs, quality gates, source-data tables, manuscript-facing figures and supplementary outputs were organised in the project repository. The analyses evaluate transportability and robustness of the predefined biological modules. They do not constitute diagnostic classifier discovery, clinical-performance validation, clinical implementation evidence or causal validation. Ethics approval This study reanalysed publicly available de-identified transcriptomic datasets and did not involve new recruitment of human participants, new collection of human biospecimens or access to identifiable private information. No new ethics approval was required for this secondary analysis of public data. Declaration of generative AI and AI-assisted technologies in the manuscript preparation process During preparation and revision of this manuscript, the author used ChatGPT, an AI-assisted tool provided by OpenAI, to assist with editorial organisation, language refinement, workflow planning, code drafting and checking, formatting checks, and preparation of manuscript and submission-support materials. AI-generated suggestions were not treated as scientific evidence. Analysis scripts were executed against the stated public datasets, and numerical and graphical outputs were checked using the documented reproducibility, source-lock and quality-control procedures. The author reviewed and verified the analysis code, results, references, biological interpretation, figures, tables, manuscript text and submission materials and takes full responsibility for the final work. Results GSE211567 discovery identifies bacterial- and viral-associated programmes across sites The GSE211567 discovery analysis used a predefined bacterial-versus-viral contrast while keeping module definition separate from external testing. The primary limma analysis ranked host-transcriptomic features before cross-site concordance checks across the pooled and site-stratified analyses (Fig 1A-B). This procedure reduced the likelihood of carrying forward features driven mainly by one geographic or technical stratum. Discovery and conservative definition of bacterial- and viral-associated host-response modules in GSE211567. (A) Primary bacterial-versus-viral discovery analysis in GSE211567. The volcano-style plot summarises the limma-ranked host-transcriptomic contrast used as the discovery starting point before cross-site concordance assessment and biological module definition. Positive log2 fold-change values indicate bacterial-higher features, whereas negative values indicate viral-higher features. (B) Cross-site concordance of the GSE211567 bacterial-versus-viral discovery contrast. Spearman correlations summarise ranked agreement of log2 fold changes between the pooled, Sri Lanka and United States analyses. Directional concordance is the percentage of compared features whose bacterial-versus-viral log2 fold changes have the same sign in the two indicated analyses. This analysis was used to prioritise signals whose direction was not restricted to a single site before pathway interpretation and module definition. (C) The five predefined GSE211567 discovery modules used in external testing. Two modules were bacterial-higher, representing cytoplasmic translation/ribosomal protein activity and mitochondrial respiration/oxidative phosphorylation. Three modules were viral-higher, representing broad antiviral/interferon-stimulated defence, viral restriction/type I interferon signalling and cytokine/innate immune regulation. Module genes and expected directions were fixed before external analysis, with no later gene reselection or module redefinition. Conservative biological curation defines five fixed modules Directionally eligible features were mapped to genes and assessed using Gene Ontology biological-process enrichment. Redundancy-reduced biological groups were reviewed using the documented curation hierarchy before final module locking. Two modules were bacterial-higher: BACT_M1, representing cytoplasmic translation and ribosomal activity, and BACT_M2, representing mitochondrial respiration and oxidative phosphorylation. Three modules were viral-higher: VIR_M1a, VIR_M1b and VIR_M2, representing broad antiviral/interferon defence, viral restriction/type I interferon activity and cytokine/innate immune regulation, respectively (Fig 1C). GSE73461 formal external projection retains all expected directions The primary GSE73461 contrast contained 52 DefiniteBacterial and 94 DefiniteViral samples. All modules passed the predefined coverage threshold and were scored without gene reselection, module redefinition, reweighting or model training. All five modules retained their discovery directions (Fig 2; Table 1). BACT_M2 was bacterial-higher, with a median bacterial-minus-viral score difference of +0.3328 and BH-adjusted Wilcoxon P = 0.0202. BACT_M1 was also bacterial-higher but remained borderline after correction, with a median difference of +0.2067 and adjusted P = 0.0799. VIR_M1a, VIR_M1b and VIR_M2 were viral-higher, with median differences of −0.4629, −0.6739 and −0.2596 and adjusted P values of 4.77 x 10^-6, 1.41 x 10^-6 and 0.00848, respectively. Primary-only z-reference scoring retained all expected directions and the same four-module pattern of adjusted statistical support. 10.1371/journal.pone.0353559.t001Table 1External evaluation of the five predefined GSE211567 discovery modules in GSE73461.ModuleBiological labelDiscoverydirectionLockedgenesGSE73461genes scoredMain projectionmedian difference;BH-adjusted PPrimary-only z-scoresensitivitymedian difference;BH-adjusted PExpected directionretained (main)Expected directionretained (sensitivity)InterpretationMissinggenesBACT_M1Bacterial-higher cytoplasmic translation and ribosomal protein programmeHigher in bacterial2524+0.2067;BH P = 0.0799+0.2211;BH P = 0.0778YesYesDirectionally concordant but borderlineHYDIN2BACT_M2Bacterial-higher mitochondrial respiration and oxidative phosphorylation programmeHigher in bacterial2121+0.3328;BH P = 0.0202+0.3504;BH P = 0.0165YesYesRobustly externally transportedVIR_M1aViral-higher broad antiviral and interferon-stimulated defence programmeHigher in viral128128−0.4629;BH P = 4.77e-06−0.4441;BH P = 7.57e-06YesYesStrongly and robustly externally transportedVIR_M1bViral-higher viral restriction and type I interferon signalling subgroupHigher in viral3333−0.6739;BH P = 1.41e-06−0.6445;BH P = 2.22e-06YesYesStrongly and robustly externally transportedVIR_M2Viral-higher cytokine and innate immune regulation programmeHigher in viral106105−0.2596;BH P = 0.00848−0.2626;BH P = 0.00796YesYesRobustly externally transportedBTN2A3PModule scores were calculated using the pre-specified unweighted mean z-score rule without gene reselection, module redefinition, reweighting or diagnostic model training. The primary projection contrast compared DefiniteBacterial and DefiniteViral samples.Discovery direction: Direction assigned during GSE211567 discovery-module locking before external projection.Locked genes: Number of genes fixed in the locked GSE211567 discovery module before GSE73461 projection.Genes scored in GSE73461: Number of locked module genes successfully mapped and scored in GSE73461 after Illumina probe annotation.Main projection result: Median bacterial-minus-viral module-score difference and Benjamini–Hochberg adjusted Wilcoxon P value from the main GSE73461 projection analysis.Primary-only z-score sensitivity result: Median bacterial-minus-viral module-score difference and Benjamini–Hochberg adjusted Wilcoxon P value after gene-wise z-scoring using only DefiniteBacterial and DefiniteViral samples, excluding Control samples from the z-score reference set.Interpretation tier: Conservative interpretation based on direction concordance, statistical support and robustness in the primary-only z-score sensitivity analysis.Abbreviations: BH, Benjamini–Hochberg; OXPHOS, oxidative phosphorylation. External evaluation of the five predefined GSE211567 discovery modules in GSE73461. Module scores were calculated using the pre-specified unweighted mean z-score rule without gene reselection, module redefinition, reweighting or diagnostic model training. The primary projection contrast compared DefiniteBacterial and DefiniteViral samples. Discovery direction: Direction assigned during GSE211567 discovery-module locking before external projection. Locked genes: Number of genes fixed in the locked GSE211567 discovery module before GSE73461 projection. Genes scored in GSE73461: Number of locked module genes successfully mapped and scored in GSE73461 after Illumina probe annotation. Main projection result: Median bacterial-minus-viral module-score difference and Benjamini–Hochberg adjusted Wilcoxon P value from the main GSE73461 projection analysis. Primary-only z-score sensitivity result: Median bacterial-minus-viral module-score difference and Benjamini–Hochberg adjusted Wilcoxon P value after gene-wise z-scoring using only DefiniteBacterial and DefiniteViral samples, excluding Control samples from the z-score reference set. Interpretation tier: Conservative interpretation based on direction concordance, statistical support and robustness in the primary-only z-score sensitivity analysis. Abbreviations: BH, Benjamini–Hochberg; OXPHOS, oxidative phosphorylation. External evaluation of predefined GSE211567 discovery modules in GSE73461. (A) Distribution of locked module scores in the independent GSE73461 external projection cohort. Module scores were calculated using the pre-specified unweighted mean z-score rule in DefiniteBacterial and DefiniteViral samples. Genes were scored without gene reselection, reweighting, module renaming or diagnostic model training. (B) Median bacterial-minus-viral module-score differences in the main GSE73461 projection and in the primary-only z-score sensitivity analysis. Positive values indicate higher scores in DefiniteBacterial samples, whereas negative values indicate higher scores in DefiniteViral samples. All five modules retained the expected discovery direction in both analyses. (C) BH-adjusted Wilcoxon P values for the main projection and the primary-only z-score sensitivity analysis are shown as independent points for each categorical module. Circles denote the main projection and triangles denote the primary-only z-score sensitivity analysis. The dashed horizontal line indicates BH-adjusted P = 0.05. The module categories are not connected because they are distinct predefined gene sets rather than points on a continuous sequence. BACT_M1 remained directionally concordant but borderline, whereas BACT_M2 and all three viral-associated modules passed the adjusted-P-value threshold in both mean-z analyses. Table 1 reports fixed-module external projection. It should not be interpreted as diagnostic classifier discovery, model training, gene rediscovery, module redefinition or causal validation. GSE72810 provides accession-level and cross-platform validation The GSE72810 primary contrast contained 23 definite bacterial and 28 definite viral samples. Coverage of the predefined module genes remained high after Entrez reconciliation and prespecified representative-probe selection. All five modules retained their expected directions. BACT_M2 had a Hodges-Lehmann bacterial-minus-viral shift of +0.425 (95% CI + 0.161 to +0.676; adjusted P = 0.0020). VIR_M1a, VIR_M1b and VIR_M2 had shifts of −0.726 (95% CI −0.907 to −0.577), −0.941 (95% CI −1.201 to −0.752) and −0.576 (95% CI −0.682 to −0.475), with adjusted P values of 7.16 x 10^-8, 8.77 x 10^-8 and 8.77 x 10^-8, respectively. BACT_M1 was bacterial-higher but borderline, with a shift of +0.266 (95% CI −0.024 to +0.722; adjusted P = 0.0799). Cross-cohort effects support four modules in both cohorts For harmonised cross-cohort comparison, Fig 3 and Table 2 report Hodges-Lehmann bacterial-minus-viral shifts with bootstrap 95% confidence intervals; these estimates are distinct from the median score differences reported for the original GSE73461 projection in Fig 2 and Table 1. All ten cohort-module effects retained the expected direction (Fig 3; Table 2). BACT_M2, VIR_M1a, VIR_M1b and VIR_M2 had confidence intervals excluding zero and passed false-discovery-rate correction in both GSE73461 and GSE72810. BACT_M1 was directionally concordant in both cohorts, but both confidence intervals included zero and both adjusted P values were approximately 0.08. 10.1371/journal.pone.0353559.t002Table 2Cross-cohort Hodges-Lehmann effects for the five locked host-response modules.ModuleGSE73461 effect(95% CI)GSE73461BH-adjusted PGSE72810 effect(95% CI)GSE72810BH-adjusted PExpected directionretained in bothCI excludes zeroin bothFDR significantin bothInterpretationBACT_M1+0.172(95% CI −0.021 to +0.377)0.0799+0.266(95% CI −0.024 to +0.722)0.0799YesNoNoDirectionally concordant in both cohorts but statistically borderline, with confidence intervals including zero.BACT_M2+0.271(95% CI + 0.057 to +0.475)0.0202+0.425(95% CI + 0.161 to +0.676)0.0020YesYesYesExpected-direction support with confidence intervals excluding zero in both cohorts.VIR_M1a−0.433(95% CI −0.617 to −0.269)4.77e-06−0.726(95% CI −0.907 to −0.577)7.16e-08YesYesYesStrong expected-direction support with confidence intervals excluding zero in both cohorts.VIR_M1b−0.574(95% CI −0.788 to −0.368)1.41e-06−0.941(95% CI −1.201 to −0.752)8.77e-08YesYesYesStrong expected-direction support with confidence intervals excluding zero in both cohorts.VIR_M2−0.228(95% CI −0.407 to −0.063)0.0085−0.576(95% CI −0.682 to −0.475)8.77e-08YesYesYesExpected-direction support with confidence intervals excluding zero in both cohorts. Cross-cohort Hodges-Lehmann effects for the five locked host-response modules. Cross-cohort validation of locked module effects in GSE73461 and GSE72810. Hodges-Lehmann bacterial-minus-viral module-score shifts are shown with stratified nonparametric bootstrap 95% confidence intervals for the five modules predefined in the GSE211567 discovery analysis. Positive estimates indicate higher module scores in bacterial samples, whereas negative estimates indicate higher scores in viral samples. GSE73461 contained 52 DefiniteBacterial and 94 DefiniteViral samples measured on GPL10558. GSE72810 contained 23 definite bacterial and 28 definite viral samples measured on GPL6947. BACT_M2, VIR_M1a, VIR_M1b and VIR_M2 retained their expected directions with confidence intervals excluding zero and BH-adjusted Wilcoxon P values below 0.05 in both cohorts. BACT_M1 retained the expected bacterial-higher direction in both cohorts but remained statistically borderline. The modules were scored without gene reselection, module redefinition, gene reweighting, direction flipping or diagnostic-model training. GSE72810 was analysed as a second accession-level and sample-level cohort providing cross-platform validation. Its GSM accession set was disjoint from GSE73461 and it used a different Illumina platform. Direct participant overlap could not be assessed because participant identifiers were not deposited, and the studies arose from the same broad investigator network. Effects are Hodges-Lehmann bacterial-minus-viral location shifts with bootstrap 95% confidence intervals. Positive values indicate bacterial-higher scores and negative values indicate viral-higher scores. Benjamini-Hochberg adjustment was applied across the five modules within each cohort. The cross-cohort pattern therefore supported four modules under both effect-size and adjusted-significance criteria. The strongest and most consistent signals were the viral-associated modules, followed by the bacterial mitochondrial respiration/OXPHOS module. Sensitivity analyses retain direction but identify method-dependent support Across the two GSE73461 z-reference analyses and four GSE72810 z-reference, case-definition and probe-collapse analyses, all 30 module estimates retained the expected direction (S1 Fig, panel A). Twenty-four estimates passed false-discovery-rate correction and 25 rank-biserial confidence intervals excluded zero. GSE72810 score representations were highly concordant; the minimum Pearson correlation was 0.9874 and the minimum Spearman correlation was 0.9814. The GSVA analysis retained all expected directions but showed non-uniform inferential support (S1 Fig, panel B). BACT_M2, VIR_M1a and VIR_M1b remained supported under both mean-z and GSVA scoring. BACT_M1 was borderline under mean-z scoring but gained confidence-interval and adjusted-P-value support under GSVA. In contrast, VIR_M2 was supported under mean-z scoring but was near zero and statistically unsupported under GSVA in both reference populations. VIR_M2 is therefore scoring-method-sensitive rather than uniformly robust across algorithms. Exhaustive deletion analysis supports distributed module signal The leave-one/two-gene analysis evaluated 29,826 module variants across the two GSE73461 scoring populations (S1 Fig, panel C). Every variant retained the expected module direction. The minimum Pearson correlation with the corresponding complete-module score was 0.9940. These results indicate that the observed module behaviour was not driven by removal-sensitive dependence on one gene or one gene pair, while not establishing causal sufficiency of individual genes. Discussion This study evaluated whether biologically curated whole-blood host-response modules discovered in GSE211567 retained their prespecified bacterial-higher or viral-higher behaviour when the same definitions and scoring rules were applied to two external cohorts. All five modules retained their expected directions in both GSE73461 and GSE72810. Four modules – BACT_M2, VIR_M1a, VIR_M1b and VIR_M2 - had confidence intervals excluding zero and passed adjusted-significance thresholds in both cohorts. BACT_M1 was directionally concordant but borderline in each cohort, indicating a reproducible direction with less stable inferential support. The viral-associated modules showed the strongest transportability. VIR_M1a and VIR_M1b captured broad antiviral, interferon-stimulated and viral-restriction programmes, while VIR_M2 represented cytokine and innate immune regulation. Their cross-cohort behaviour is consistent with the central role of interferon-linked responses in viral infection and with prior studies in which interferon-inducible biomarkers contributed to viral-versus-bacterial discrimination [3,4,19]. BACT_M2 also transported consistently, supporting a bacterial-associated mitochondrial respiration and oxidative-phosphorylation programme in these datasets and aligning with evidence that inflammatory activation can remodel leukocyte immunometabolism [20,21]. The sensitivity analyses refine rather than uniformly strengthen this interpretation. Direction was stable across z-reference populations, GSE72810 case definitions and probe handling, and GSE72810 score correlations remained high. Exhaustive deletion analysis also supported a distributed module signal. However, GSVA changed inferential support for two modules: BACT_M1 became supported, whereas VIR_M2 lost statistical support despite retaining the expected sign. Directional robustness, effect-size robustness and statistical robustness should therefore be reported as related but distinct properties. The principal contribution is not a diagnostic classifier, but a transparent framework for asking whether predefined biological modules retain their behaviour in new datasets. Fixing the module genes, expected directions and scoring rules before external testing, and prohibiting gene reselection, sign reversal, reweighting and model retraining, reduces rediscovery bias. Hodges-Lehmann shifts, rank-biserial effects, confidence intervals, cross-platform testing and deletion analyses provide complementary evidence about the magnitude and stability of the observed programmes. Several limitations remain. Public metadata and infection adjudication were imperfect, and the cohorts differed in age, syndrome, pathogen composition, platform and preprocessing. Whole-blood scores may reflect both cell-composition changes and cell-intrinsic activation. The GSE72810 and GSE73461 GEO sample accession sets were disjoint and were measured on different Illumina platforms, but direct participant overlap could not be assessed because participant identifiers were not deposited; the studies also arose from the same broad investigator network. GSE72810 should therefore not be treated as a fully investigator-independent replication cohort. The retrospective analyses do not establish clinical performance, clinical readiness or causal mechanisms. Prospective, investigator-independent and multi-cohort studies are required to determine how these modules behave across age groups, syndromes, pathogens, sampling times and treatment contexts. Transparency declaration Code availability. Analysis scripts, decision logs, quality gates, source-data tables, manuscript-facing figures and supplementary outputs are available in the public project repository at https://github.com/rmaghembe1/host-response-module-transportability. The revision-round code and outputs used for this resubmission are available on the public plosone_revision_round1_2026 branch at https://github.com/rmaghembe1/host-response-module-transportability/tree/plosone_revision_round1_2026. Supporting information Supplementary Tables S1–S5. External-cohort search register, locked GSE211567 module definitions, GSE73461 identifier coverage and probe choices, GSE73461 sample-level scores, and complete GSE73461 primary and z-reference sensitivity statistics. (XLSX) Supplementary Table S6. GSE72810 cohort audit, sample-classification framework, locked primary and expanded contrasts, and cross-cohort independence assessment. (XLSX) Supplementary Table S7. GSE72810 Entrez reconciliation, module coverage, complete module-gene mapping, and prespecified representative-probe choices. (XLSX) Supplementary Table S8. Privacy-safe GSE72810 sample-level module scores using anonymized analysis identifiers, primary and sensitivity tests, bootstrap effect-size confidence intervals, and score-concordance analyses. (XLSX) Supplementary Table S9. Harmonised GSE73461–GSE72810 cross-cohort effect-size source data and five-module summary. (XLSX) Supplementary Table S10. GSE73461 mean-z/GSVA comparison, score concordance, module-level deletion summary, worst-case deletion variants, and complete exhaustive leave-one/two-gene results. (XLSX) Sensitivity and robustness of locked host-response modules. A, Z-reference, case-definition and probe-collapse sensitivities. Rank-biserial bacterial-versus-viral effects and bootstrap 95% confidence intervals are shown for the two GSE73461 mean-z analyses and four GSE72810 analyses. All 30 cohort-module estimates retained the expected direction. GSE72810 score concordance across the tested z-reference and probe-collapse representations remained high, with minimum Pearson r = 0.9874 and minimum Spearman rho = 0.9814. B, Mean-z versus GSVA scoring-method sensitivity in GSE73461. Rank-biserial effects are displayed because they are comparable across scoring methods with different raw score scales. BACT_M2, VIR_M1a and VIR_M1b retained confidence-interval and BH-adjusted statistical support under both methods and z-reference populations. BACT_M1 was borderline under mean-z scoring but supported under GSVA. VIR_M2 retained a small viral-higher direction under GSVA but its confidence intervals included zero and its BH-adjusted P values were not significant. VIR_M2 is therefore described as scoring-method-sensitive. C, Exhaustive leave-one/two-gene robustness in GSE73461. Points show the minimum Pearson correlation between each deletion variant and the corresponding complete-module score. Across 29,826 variants, every leave-one and leave-two analysis retained the expected module direction. The minimum Pearson correlation was 0.9940 and the minimum Spearman correlation was 0.9897. Positive rank-biserial effects indicate bacterial-higher scores and negative effects indicate viral-higher scores. These analyses evaluate robustness of the predefined modules and do not constitute gene reselection, module redefinition, diagnostic-model training or causal validation. (PNG) 20 Jul 2026 Dear Dr. Maghembe, 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 Sep 03 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, Guichuan Lai 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. Thank you for your submission to PLOS ONE. Before we can proceed, we kindly ask you to address the following concerns: We note that you have declared the use of generative AI in the preparation of your manuscript. Please note that contributions by artificial intelligence (AI) tools and technologies to a study or to an article’s contents must be clearly reported in a dedicated section of the Methods, or in the Acknowledgements section for article types lacking a Methods section. This section should include the name(s) of any tools used, a description of how the authors used the tool(s) and evaluated the validity of the tool’s outputs, and a clear statement of which aspects of the study, article contents, data, or supporting files were affected/generated by AI tool usage (https://journals.plos.org/plosone/s/ethical-publishing-practice#loc-artificial-intelligence-tools-and-technologies). Please therefore revise your manuscript to provide all necessary information to meet these requirements. We appreciate your attention to these queries and look forward to your response. 3. 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, all author-generated code must 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 4. Please note that your Data Availability Statement is currently missing a direct link to access each database. If your manuscript is accepted for publication, you will be asked to provide these details on a very short timeline. We therefore suggest that you provide this information now, though we will not hold up the peer review process if you are unable. 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. [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: No Reviewer #2: Yes ********** 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: No ********** Reviewer #1: This manuscript investigates consistence of host-response transcriptomic associated with bacterial and viral infections. The author proposes a conservative workflow based on modules and their transportability in an external dataset. Overall, the distinction between discovery and external projection is well maintained. - The main limitation of the study is that the conclusions are based on only one external validation dataset (GSE73461). Include additional datasets or discuss this limitation more explicitly. For example: "These results support the external transportability of selected host-response modules across public infection transcriptomic cohorts." This statement could be slightly softened to avoid overgeneralization. - The manuscript indicates that Gene Ontology terms were manually reviewed to define modules. More details (criteria) describing how these decisions were made would improve transparency and reproducibility. Are these modules optimal or most informative? - The manuscript does not compare the proposed strategy with other commonly pathway scoring approaches (GSVA). A brief discussion explaining these limitations should be addressed. Only median differences and adjusted P-values are presented. Include effect size measures and confidence intervals. Overall, the manuscript addresses a relevant methodological question regarding the reproducibility and transportability of host-response transcriptomic modules across independent datasets. The novelty is methodological rather than biological, regarding the reproducibility of biologically interpretable transcriptomic modules across independent cohorts. Reviewer #2: SYNOPSIS AND TECHNICAL ISSUES This manuscript describes a limited and largely sound piece of research seeking to identify characteristic patterns of gene expression that differentiate a viral infection from a bacterial infection in humans. The work explores transcript levels from two published compendiums of transcript information with known disease status. The premise of this work is that there exists collections of genes for which the change in expression reflects the infecting agent and that such collections can be applied more broadly. As I understood the report, one set of data was used as the training set to extract sets of characteristic genes. The expression of these sets of genes was analyzed in a second set of published data to evaluate the discriminatory power of the collection. The authors conclude that it is possible to identify such collections of genes that could be "transported" to other data sets. The study was limited to these two data sets and did not aggressive explore the power of these particular findings. Specifically, the collections of genes were identified by computational means and with guidance provided to the investigator using information about biological processes (GO identifiers.) The premise was supported by the findings but not challenged. For example, how many of the identified genes were essential for each collection to provide discrimination? This would have been a valuable computational exercise (leave one out, leave two out, etc.) that could extend the utility of the genes identified in this single training set. The singularly significant finding reported in the Discussion was that discrete programs of gene expression could be found with the power to discriminate among the infecting agent. Viral infections were characterized by antiviral pathways, interferon-stimulated pathways and the more vague anti-viral program of gene expression. Bacterial disease appeared to have a curious fingerprint in translational and ribosome metabolism along with mitochondrial respiration and oxidative phosphorylation. This finding may prove to be of greater significance in future studies that violate the "directionally" found in these studies. For example, a novel viral disease that presents with the pattern of the bacterial fingerprint may provide unexpected insight into a novel nature of the viral disease and vice-versa. It is for this reason I find the study to be valuable. However, there are additional significant limitations in the manner in which the work was presented. Technical limitations in this manuscript that need to be corrected include the following: 1. As noted above, the extent to which the gene sets can suffer the loss of specific members and still provide discrimination should be evaluated. Alternatively, if my understanding is so badly off-base, the manuscript should be re-written in a manner to make the objectives more clear to a naïve reader such as myself. 2. The gene sets analyzed were only identified by the accession number (GSE211567 and GSE73461). The nature of the samples needs to be described in substantial detail for independent evaluation by the reader. The content and nature of the gene sets matter much more than the accession number. 3. What was the logic of excluding control samples in the z-score reference set (lines 118-119). Specifically, what was the expected outcome of this exercise and how could it have affected the results? 4. The figure legends (captions) are entirely inadequate because they fail to explain what is depicted in each figure. These legends should state the finding, succinctly describe the method of analysis, and describe the elements in the figure needed to ascertain the finding. 5. What is the formal definition of concordance used here as shown in Fig. 1B? 6. What is the formal definition and mode of calculation for the "module score" from the mean z-score. What is this mean value derived from? 7. What is shown in Fig. 2C? Because the x-axis is not a continuum of values, the y-values have little reason to be joined and expressed in a scatter plot as shown. Moreover, I am a bit unsure of how one defines "projection significance" and what is meant by the "Main projection and "Primary-only z-score sensitivity? STATISTICS It is difficult to evaluate the statistical analyses because the methods were not adequately (or clearly) described. In so far as the primary query was a z-score, the results appear appropriate analyzed using corrections for multiple comparisons. CLARITY The manuscript is very difficult to understand because of the overly generous use of jargon and phrases that have minimal informative content. I would strongly encourage the authors to write in a simpler style. Even the word "transportable" leans toward jargon. This word, in the context of bioinformatic analyses, should be defined in plain English before being used so heavily. Other phrases that are unnecessarily confusing include: "coherent expected-direction behavior," "site-aware discovery," "safeguards" (in the context of the analysis of gene expression), "strict discovery/projection firewall," and "frozen," to highlight phrases from the Introduction that could better be conveyed in simple English. Of particular importance is the notion of "projection." The focus of this paper seems to be that clusters of related genes can be identified and applied to a new dataset to discriminate between viral and bacterial disease without redefining (or rediscovering) the clusters of related genes. This method of "fixed-module projection" should be defined in plain English to reach the broader audience that is typical for the PLoS journals. ********** 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.] 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. 10 Aug 2026 Dear Academic Editor and Reviewers, Thank you for the careful evaluation of the manuscript and for the constructive recommendations. The manuscript has been revised substantially in response. The principal changes include addition of a second accession-level and cross-platform projection cohort (GSE72810), harmonised effect-size and confidence-interval reporting, GSVA sensitivity analysis, exhaustive leave-one/two-gene robustness analysis, expanded cohort and statistical-method descriptions, a revised categorical Figure 2C, clearer definitions and simpler terminology, direct public-data links, public code availability, and a dedicated generative-AI disclosure within Methods. Each comment is reproduced below, followed by the response and the location of the corresponding revision. The revised submission includes separate clean and marked-up manuscript files. Academic Editor / Journal Requirements J1. PLOS ONE style and revision-package requirements Comment: Ensure PLOS ONE style requirements and file naming are followed. Response: Thank you. The revision package has now been prepared in PLOS ONE submission form. Separate clean and marked-up manuscript files have been generated from the same locked scientific source and subjected to structural, rendering and visual quality-control checks. Continuous line numbering is retained, the two wide tables remain editable and readable in landscape sections, and the marked-up file uses genuine tracked revisions while accepting those revisions reproduces the clean manuscript content. Changes in the revised manuscript: Final revision package: PONE-D-26-30583_clean_revised_manuscript.docx and PONE-D-26-30583_marked_up_revised_manuscript.docx. Supporting analysis/material: Clean manuscript and marked-up manuscript passed the final DOCX structural and visual QA gates. J2. Generative-AI disclosure Comment: Provide a dedicated generative-AI disclosure naming the tool, describing use, validation of outputs, and affected study/article materials. Response: The manuscript now contains a dedicated generative-AI disclosure within the Methods section. The disclosure identifies ChatGPT as an AI-assisted tool provided by OpenAI, describes its use for editorial organisation, language refinement, workflow planning, code drafting and checking, formatting checks and preparation of submission-support materials, states that AI-generated suggestions were not treated as scientific evidence, describes verification of numerical and graphical outputs, and states that the author reviewed the analysis, references, interpretation, figures, tables and manuscript and takes responsibility for the final work. Changes in the revised manuscript: Methods, 'Declaration of generative AI and AI-assisted technologies in the manuscript preparation process', immediately before the main Results section. Supporting analysis/material: The declaration wording was preserved during relocation into Methods; clean and marked-up DOCX copies were synchronised and visually checked. J3. Public availability of author-generated code Comment: Make author-generated code publicly available without restriction. Response: The author-generated analysis code and reproducibility materials are publicly available without access restriction in the project repository. The revision branch contains the revision-round scripts, decision logs, quality gates, source-data tables, manuscript-facing figures and supplementary outputs used to support the revised analyses. Changes in the revised manuscript: Code availability section. Supporting analysis/material: Public repository: https://github.com/rmaghembe1/host-response-module-transportability; revision branch: https://github.com/rmaghembe1/host-response-module-transportability/tree/plosone_revision_round1_2026. J4. Direct database links in Data Availability Comment: Provide direct links to each database in the Data Availability Statement. Response: The Data Availability Statement has been revised to provide direct NCBI Gene Expression Omnibus links for each of the three principal datasets: GSE211567, GSE73461 and GSE72810. The text also identifies candidate or technical-rehearsal datasets considered during workflow development. Changes in the revised manuscript: Data availability section. Supporting analysis/material: Direct GEO accession URLs for GSE211567, GSE73461 and GSE72810 are included in the revised manuscript. Reviewer 1 Reviewer 1, Comment 1 — Additional external cohort and limitation Comment: The conclusions relied on one external validation dataset; include additional datasets or discuss/soften this limitation. Response: We agree that reliance on a single projection cohort was an important limitation. We therefore added GSE72810 as a second accession-level and sample-level whole-blood cohort measured on a different Illumina platform (GPL6947). The cohort contains 146 samples; the prespecified primary contrast comprises 23 definite bacterial and 28 definite viral samples. All five modules retained their expected directions. BACT_M2, VIR_M1a, VIR_M1b and VIR_M2 had confidence intervals excluding zero and passed false-discovery-rate correction in both GSE73461 and GSE72810, whereas BACT_M1 remained directionally concordant but borderline in both. We also softened the replication language because the two GEO accession sets were disjoint and used different array platforms, but direct participant overlap could not be assessed and the studies arose from the same broad investigator network. Changes in the revised manuscript: Methods: 'GSE72810 cross-platform validation and probe selection' and 'Cross-cohort interpretation and reproducibility boundaries'; Results: 'GSE72810 provides accession-level and cross-platform validation' and 'Cross-cohort effects support four modules in both cohorts'; Discussion limitations. Supporting analysis/material: Figure 3; Table 2; Supplementary Tables S6-S9; GSE72810 primary Hodges-Lehmann shifts and bootstrap confidence intervals. Reviewer 1, Comment 2 — Manual GO review and module definition Comment: Provide more detail on criteria used during manual GO review/module definition and clarify whether the modules are optimal or most informative. Response: We expanded the Methods to make the module-definition procedure more transparent. Directionally supported discovery features were mapped to genes, assessed by Gene Ontology biological-process enrichment, and reviewed after grouping overlapping or redundant enriched terms. Candidate programmes were evaluated using the documented biological curation hierarchy. We now state explicitly that this was a biologically guided, reproducible curation step rather than mathematical optimisation for group separation or prediction accuracy. External-cohort results were not used to choose or modify module genes. Changes in the revised manuscript: Methods, 'GSE211567 discovery and module definition'. Supporting analysis/material: Locked module definitions are unchanged; the revision clarifies the selection rationale and its non-optimisation character. Reviewer 1, Comment 3 — Comparison with GSVA Comment: Compare the proposed scoring strategy with GSVA. Response: We added a GSVA sensitivity analysis in GSE73461 using the same unchanged module gene sets and both z-reference populations. BACT_M2, VIR_M1a and VIR_M1b retained statistical support under both mean-z and GSVA scoring. BACT_M1 was borderline under mean-z scoring but gained statistical support under GSVA. VIR_M2 retained its expected viral-higher direction but became near-zero and statistically unsupported under GSVA. The revision therefore distinguishes direction preservation from method-independent inferential support and describes VIR_M2 as scoring-method-sensitive. Changes in the revised manuscript: Methods, 'Sensitivity and robustness analyses'; Results, 'Sensitivity analyses retain direction but identify method-dependent support'; Discussion. Supporting analysis/material: S1 Fig panel B and Supplementary Table S10. Reviewer 1, Comment 4 — Effect sizes and confidence intervals Comment: Report effect sizes and confidence intervals in addition to medians and adjusted P values. Response: We added effect-size estimates and uncertainty intervals throughout the external-cohort analyses. The revised Methods specify Hodges-Lehmann location shifts, rank-biserial effects and stratified 10,000-replicate bootstrap 95% confidence intervals. Figure 3 and Table 2 provide harmonised cross-cohort Hodges-Lehmann effects and 95% confidence intervals for all five modules. This allows the magnitude and uncertainty of the observed effects to be assessed separately from adjusted P values. Changes in the revised manuscript: Methods, 'Module scoring and statistical analysis'; Results, GSE72810 and cross-cohort effect sections; Figure 3; Table 2. Supporting analysis/material: Supplementary Tables S8-S9 and cross-cohort Figure 3 source data. Reviewer 2 Reviewer 2, Comment 1 — Leave-one/two-gene robustness Comment: Evaluate how loss of specific genes affects module discrimination, for example leave-one-out and leave-two-out analyses. Response: We performed exhaustive leave-one-gene and leave-two-gene sensitivity analyses for every module under both GSE73461 scoring-reference populations. In total, 29,826 module variants were evaluated. Every variant retained the expected module direction. The minimum Pearson correlation between a deletion variant and its corresponding complete module score was 0.9940, and the minimum Spearman correlation was 0.9897. These results support a distributed module signal rather than dependence on removal-sensitive individual genes or gene pairs. Changes in the revised manuscript: Methods, 'Sensitivity and robustness analyses'; Results, 'Exhaustive deletion analysis supports distributed module signal'. Supporting analysis/material: S1 Fig panel C; Supplementary Table S10, including complete leave-one/two-gene results. Reviewer 2, Comment 2 — Dataset description Comment: Describe the nature and sample composition of the analyzed datasets in substantially greater detail. Response: The dataset descriptions have been substantially expanded. GSE211567 is identified as whole-blood transcriptomic data from Sri Lanka and the United States, with a prespecified discovery set of 224 samples (101 bacterial and 123 viral). GSE73461 is described with its GPL10558 platform, 52 definite bacterial, 94 definite viral and 55 control samples in the main z-score reference population, and the restricted primary inferential contrast. GSE72810 is described as 146 paediatric whole-blood samples on GPL6947, including definite, probable, control and uncertain groups and the prespecified 23-versus-28 primary contrast. Changes in the revised manuscript: Methods, 'Study design and datasets', 'GSE211567 discovery and module definition', 'GSE73461 formal external projection', and 'GSE72810 cross-platform validation and probe selection'. Supporting analysis/material: Supplementary cohort-audit and sample-classification tables. Reviewer 2, Comment 3 — Control samples and z-score reference Comment: Explain the logic and consequences of control-sample inclusion/exclusion in the z-score reference population. Response: We clarified the role of control and non-primary samples in score standardisation versus inference. In GSE73461, the 52 definite bacterial, 94 definite viral and 55 control samples constitute the main gene-wise z-score reference population, but control samples are not included in the bacterial-versus-viral inferential test. A primary-only z-reference sensitivity analysis quantifies dependence on that choice. In GSE72810, all 146 samples contribute to the main z-score reference population, whereas the primary inferential contrast remains restricted to the 23 definite bacterial and 28 definite viral samples. Additional case-definition and probe-collapse analyses test the robustness of these choices. Changes in the revised manuscript: Methods, 'GSE73461 formal external projection', 'GSE72810 cross-platform validation and probe selection', and 'Sensitivity and robustness analyses'. Supporting analysis/material: Figure 2, S1 Fig panel A and the supplementary sensitivity tables. Reviewer 2, Comment 4 — Figure captions Comment: Rewrite figure captions so they explain the finding, analysis and figure elements. Response: All principal figure legends were rewritten so that they explain both the analysis and the graphical elements. The revised legends define direction conventions, sample groups, symbols, confidence intervals or adjusted-P-value thresholds where applicable, and state the principal finding represented in each panel. The S1 Fig legend likewise explains the z-reference/case-definition/probe-collapse sensitivity analyses, mean-z versus GSVA comparison and exhaustive gene-deletion analysis. Changes in the revised manuscript: Legends for Figures 1-3 and Supporting Information S1 Fig. Supporting analysis/material: Revised manuscript figure captions and supporting-information caption. Reviewer 2, Comment 5 — Formal concordance definition Comment: Provide the formal definition of concordance used in Figure 1B. Response: We now define directional concordance formally. For each modelled feature, concordance means agreement in the sign of the bacterial-versus-viral log2 fold change between the two contrasts being compared. Percentage directional concordance is the number of same-sign features divided by the number of compared features and multiplied by 100. Spearman correlations of log2 fold changes are reported separately as a complementary measure of ranked effect-size agreement. Changes in the revised manuscript: Methods, 'GSE211567 discovery and module definition'; Figure 1B legend. Supporting analysis/material: The definition now matches the implemented cross-site concordance calculation. Reviewer 2, Comment 6 — Module-score definition Comment: Define the module score and explain what the mean z-score represents. Response: We added the explicit module-scoring equations. For mapped gene g in sample i, expression is standardised within the prespecified reference population as z_gi = (x_gi - mean_g) / SD_g. For a module with K mapped and available genes, the sample-level score is the unweighted arithmetic mean, score_i = (1/K) sum_g z_gi. Thus each available mapped gene contributes equally to the primary score; missing genes are omitted only after module coverage has been documented. Changes in the revised manuscript: Methods, 'Module scoring and statistical analysis'. Supporting analysis/material: The explicit equations and interpretation of the mean z-score are now given in the manuscript. Reviewer 2, Comment 7 — Figure 2C categorical display Comment: Clarify Figure 2C and remove inappropriate lines connecting categorical module values. Response: Figure 2C was rebuilt to treat the module categories correctly as independent categorical values. Circles denote the main projection and triangles denote the primary-only z-score sensitivity analysis. The module categories are not connected by lines. The dashed horizontal line marks BH-adjusted P = 0.05. The revised caption explains each symbol and why no connecting line is used. Changes in the revised manuscript: Figure 2C and Figure 2 legend. Supporting analysis/material: Revised Figure 2C generated by the reproducible figure script; publication outputs include PNG and editable SVG versions. Reviewer 2 — Statistical-analysis description Comment: Improve description of statistical analyses. Response: The statistical-analysis description has been expanded substantially. The revised Methods specify Wilcoxon rank-sum testing, Benjamini-Hochberg correction across the five modules, the interpretation of positive and negative effects, median bacterial-minus-viral differences, Hodges-Lehmann shifts, rank-biserial effects, 10,000-replicate bootstrap confidence intervals, Pearson and Spearman score correlations, and the interpretive basis for cross-method comparisons in the GSVA sensitivity analysis. Changes in the revised manuscript: Methods, 'Module scoring and statistical analysis' and 'Sensitivity and robustness analyses'. Supporting analysis/material: Supplementary statistical outputs provide the corresponding complete module-l Submitted filename: Response_R1.docx 26 Aug 2026 External transportability of bacterial- and viral-associated host-response modules across public transcriptomic cohorts PONE-D-26-30583R1 Dear Dr. Maghembe, 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, Benjamin M. Liu, PhD, D(ABMM), MB(ASCP) Academic Editor PLOS One Additional Editor Comments (optional): Reviewers' comments: Reviewer's Responses to Questions Comments to the Author Reviewer #1: All comments have been addressed Reviewer #2: All comments have been addressed ********** 2. Is the manuscript technically sound, and do the data support the conclusions??> Reviewer #1: Yes Reviewer #2: Yes ********** 3. Has the statistical analysis been performed appropriately and rigorously? -->?> Reviewer #1: Yes Reviewer #2: Yes ********** 4. Have the authors made all data underlying the findings in their manuscript fully available??> The PLOS Data policy Reviewer #1: Yes Reviewer #2: Yes ********** 5. Is the manuscript presented in an intelligible fashion and written in standard English??> Reviewer #1: Yes Reviewer #2: Yes ********** Reviewer #1: Main concerns have been satisfactorily addressed as the inclusion of GSE72810 and GSVA sensitivity analysis. A few minor suggestion to interpretation of VIR_M2: consistency of VIR_M2 across cohorts was not supported maintained with GSVA. Repetition: overstatement the absence of gene reselection, module redefinition, reweighting, or model training. Reviewer #2: All concerns (of reviewer 2) have been well addressed. The reduction in jargon appears to have made the manuscript more accessible to the non-expert reader. ********** 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: Cezar Rangel Pestana Reviewer #2: No ********** PONE-D-26-30583R1 PLOS One Dear Dr. Maghembe, 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. Benjamin M. Liu Academic Editor PLOS One