Exploring the role of PANoptosis in idiopathic pulmonary fibrosis based on scRNA-seq and bulk-seq Authors: Zhihua Wang, Quanlei Li, Yuntian Chen, Lixing Gan, Lifen Yuan, Juan Liu, Yu Xie, Tianyu Zhou, Xiahui Ge Published: 2025 ABSTRACT AbstractPANoptosis is a novel form of programmed cell death that integrates pyroptosis, apoptosis, and necroptosis; in this study, we combined the single-cell RNA sequencing (scRNA-seq) dataset GSE227136 with transcriptome data to elucidate its role in idiopathic pulmonary fibrosis (IPF) pathogenesis. PANoptosis-related genes were compiled from GeneCards and published literature. Consensus clustering was used to identify distinct PANoptosis-related clusters of IPF in the GEO dataset based on filtered PANoptosis-related differentially expressed genes (PRDEGs). Specific hub genes were identified using weighted gene co-expression network analysis (WGCNA) and two machine learning methodologies, which were used to develop predictive models. The inflammatory programmed cell death score (PANoptosis score, Ps) for each IPF patient was calculated based on nine PRDEGs, followed by analyses of these PRDEGs’ expression differences and their ROC curves. PRDEG expression was confirmed in murine pulmonary tissues using quantitative real-time polymerase chain reaction (qRT-PCR). We successfully identified nine PRDEGs and two distinct PANoptosis-related clusters with these PRDEGs. Using WGCNA and machine learning approaches, we constructed a nomogram with robust predictive capacity for diagnosis of IPF. In addition, immune infiltration analysis among different molecular groups and single cell analysis revealed that increased PANoptosis activity was closely associated with immune activation. Finally, results from qRT-PCR showed a significant increase in the expression of MLKL and AIM2 in the lung tissue of the IPF animal model.Supplementary InformationThe online version contains supplementary material available at 10.1007/s10238-025-01990-8.Keywords: Idiopathic pulmonary fibrosis, PANoptosis, PANoptosis score, Single-cell RNA sequencing, Prediction model, Immune infiltration Introduction Idiopathic pulmonary fibrosis (IPF) is a chronic, progressive interstitial lung disease characterized by destruction of the alveolar architecture and progressive decline in lung function [1, 2]. The prevalence of IPF ranges from 7 to 42 per 100,000 individuals, depending on the case definition [3]. The etiology of IPF remains unclear; however, environmental factors, genetic predisposition, and age are thought to have an impact on its development [4–7]. Given the complex and multifactorial nature of IPF, there is a critical need for innovative biomarkers that facilitate early diagnosis and personalized treatment approaches. Recent investigations have highlighted various cellular mechanisms involved in the development of IPF, including abnormal epithelial cell injury, impaired repair processes, and dysregulated inflammatory responses [8, 9]. PANoptosis is a novel form of programmed cell death that integrates pyroptosis, apoptosis, and necroptosis, and has recently become a major focus of research [10, 11]. PANoptosis is associated with various pulmonary disorders, such as smoking-related emphysema, acute respiratory distress syndrome, and coronavirus disease 2019 [12–15]. PYCARD, involved in the PANoptosis pathway, may serve as a reliable biomarker for the diagnosis and treatment of chronic obstructive pulmonary disease [16]. Ma et al. reported that PM2.5 exposure hinders PANoptosis in cells and contributes to lung cancer development by affecting the EIF4B/WTAP/m6A axis [17]. Nevertheless, the intricate interactions between various cell death mechanisms and the fibrotic response require further investigation to clarify their roles in disease progression and to explore potential therapeutic avenues. Therefore, we systematically investigated the involvement of PANoptosis-related genes (PRGs) in IPF using a comprehensive bioinformatics approach. This exploration yields significant insights into the fibrotic processes and identifies potential biomarkers for the early diagnosis of IPF. Materials and methods Acquisition of transcriptomic data We retrieved the GSE10667 [18], GSE24206 [19], and GSE53845 [20] datasets from the Gene Expression Omnibus (GEO) database (www.ncbi.nlm.nih.gov/geo). The GSE10667 (GPL4133) dataset comprised samples from Homo sapiens, including eight IPF samples and 15 normal samples; the GSE24206 (GPL570) dataset al.so comprised samples from Homo sapiens, including 17 IPF samples and six normal samples; and the GSE53845 (GPL6480) dataset contained 40 IPF samples and eight normal samples. We obtained 27 PRGs from the GeneCards database using “PANoptosis” as the search term [21]. We also identified 43 PRGs in a search of PubMed using “PANoptosis” as the search term [22]. After integrating the findings and removing duplicates, we compiled a final list of 41 PRGs, as detailed in Supplementary Table 1. Identification of differentially expressed genes (DEGs) associated with pulmonary fibrosis To elucidate the mechanisms and biological features of the DEGs in IPF, we initially merged the GSE10667, GSE24206, and GSE53845 datasets, and then adjusted for batch effects using the R package sva. The combined datasets were normalized to create the pulmonary fibrosis dataset. Subsequently, we conducted differential expression analysis of the GEO dataset using the limma package to detect DEGs between the normal and IPF group [23]. Next, we filtered the genes for further analysis based on |logFC| > 0 and p < 0.05. Specifically, genes with logFC > 0 and p < 0.05 were classified as upregulated DEGs, and those with logFC < 0 and p < 0.05 were categorized as downregulated DEGs. We identified PANoptosis-related differentially expressed genes (PRDEGs) associated with IPF by intersecting the DEGs with the PRGs, and illustrated the overlapping genes using a Venn diagram. The outcomes of the differential analysis were visualized in a volcano plot, constructed with the R package ggplot2, and a heatmap for the PRDEGs was generated with the R package pheatmap. Functional enrichment analysis Gene Ontology (GO) analysis was employed to perform extensive functional enrichment evaluations of biological processes, molecular functions, and cellular components. Kyoto Encyclopedia of Genes and Genomes (KEGG) analysis was used to obtain detailed information on genomics, biological pathways, pathologies, and pharmacology. Gene set enrichment analysis (GSEA) was used to assess the enrichment profiles of gene sets in an ordered list of genes associated with specific phenotypes. For the analysis, we obtained the “c2.cp.v7.2.symbols.gmt” gene set from the Molecular Signatures Database (MSigDB). We utilized the clusterProfiler package to conduct enrichment analysis [24], applying selection threshold with p < 0.05 and false discovery rate (FDR) < 0.05. The Benjamini-Hochberg procedure was used to adjust the P-values for this analysis. PANoptosis-centered molecular grouping (clusters & risk groups) of IPF disease samples Consensus clustering is a resampling technique designed to delineate individual samples and their respective subgroup classifications [25]. The ConsensusClusterPlus package (version 1.64.0) was used to identify distinct PANoptosis-related molecular clusters of IPF in the GEO dataset based on filtered PRDEGs. Additionally, we utilized the ssGSEA algorithm to compute PANoptosis scores derived from the PRDEGs for 65 IPF patients and then divided them into high- and low-risk group. The PRDEGs expression profiles across different molecular groups were evaluated via the Mann-Whitney U test, establishing a significance level at p < 0.05. Construction of weighted gene co-expression network analysis (WGCNA) WGCNA was used to identify specific clusters and co-expression modules in a dataset [26]. The analysis selects a soft threshold and constructs a topological overlap matrix. Genes in the top 50% of the variance were selected for subsequent WGCNA analyses. The module-trait correlations were analyzed, which helped to pinpoint specific gene expression modules that exhibited a strong association with the phenotypes. We performed WGCNA analysis separately on the entire GEO dataset and the PANoptosis-related molecular clusters. The results of the two analyses were combined to identify 18 key predictive genes for further investigation. Machine learning models A support vector machine (SVM) is a supervised learning algorithm used for classification and regression analysis, particularly to process high-dimensional data [27]. Conversely, random forest (RF) ensemble learning is based on decision trees and enhances prediction accuracy and model stability by introducing randomness into the training process [28]. Using the “caret” R package, we developed both the machine learning algorithms based on the expression profiles of key genes. Finally, we selected the top ten genes from each method and identified common genes, yielding six genes as key predictive genes for IPF. Construction and validation of the predictive model A nomogram is a graphical predictive model that analyzes multiple quantitative and qualitative variables to predict a specific event. Each variable in the multifactor regression model received a score that cumulatively estimated the likelihood of an event occurring. To explore the clinical diagnostic implications of PRDEGs in conjunction with the six key predictive genes associated with IPF, we used the “rms” R package to develop the nomogram. Subsequently, we used decision curve analysis (DCA) and a calibration curve to assess the accuracy and discrimination of the logistic regression model. Furthermore, we applied a separate dataset GSE53845 to demonstrate the predictive power of the model and presented graphically using the “pROC” R package. Gene set variation analysis (GSVA) GSVA is a non-parametric and unsupervised method that evaluates the enrichment results of gene sets across various samples by converting gene expression data into a gene set expression profile [29]. The proposed approach enabled the assessment of pathway enrichment across various sample sets. For our analysis, we acquired the “h.all.v7.4.symbols.gmt” gene set from the MSigDB repository to perform GSVA, which allowed us to determine differences in functional enrichment between the normal and IPF group. We also employed the “c2.cp.kegg. v2023.1.Hs. symbols” and “c5.go. v2023.1.Hs. symbols” datasets for additional pathway enrichment analysis. Utilizing the limma package, we compared the GSVA scores of the PANoptosis-related clusters to identify biological functions and expression pathways. GSVA scores that had t-values above 2 were deemed statistically significant. Immune infiltration analysis CIBERSORT (https://cibersort.stanford.edu/) is an application that employs linear support vector regression to unravel the expression profiles of various human immune cell subtypes [30]. The CIBERSORT algorithm was used to evaluate immune cell infiltration in the GEO dataset. We also used the single-sample GSEA (ssGSEA) algorithm from the GSVA R package to calculate enrichment scores indicative of the infiltration levels of different immune cell types [31]. Spearman’s correlation analysis was utilized to investigate the interaction between immune-infiltrating cells and specific genes. Subsequently, we selected immune cells with significant differences (p < 0.05) to construct a heatmap illustrating the correlations between these immune cell types. Single cell analysis The GSE227136 dataset, which comprises 67 patients with pulmonary fibrosis and 49 unaffected donors, was used to perform cell clustering and cell type identification, with Seurat (version 5.0) utilized as the analytical tool [32]. First, the NormalizeData function was employed to standardize the data, followed by scaling with the ScaleData function to eliminate differences in expression levels between genes. After data preprocessing, the FindNeighbors and FindClusters functions with settings of dim = 1:30 and resolution = 0.8 were used to select the first 30 principal components for clustering. Uniform Manifold approximation and Projection (UMAP) was applied for cell type visualization. The expression of PRDEGs in different cells was displayed using the “pheatmap” R package. Finally, we utilized the AUCell_calcAUC function to calculate the PANoptosis score of each cell and analyzed the average PANoptosis activity within different cell types. Construction of mRNA-miRNA, mRNA-RNA-binding proteins (RBPs), mRNA-transcription factors (TFs), and mRNA-drugs interaction networks The miRTarBase database (https://mirtarbase.cuhk.edu.cn/) was used to collect experimental evidence on miRNA-mRNA targeting relationships [33]. The miRDB database (https://mirdb.org/mirdb) also predicts miRNA target genes [34]. We predicted miRNAs interacting with hub genes using both databases, and then combined the results to construct an mRNA-miRNA interaction network. Similarly, the ENCORI database was used to identify RBPs that were correlated with hub genes (https://starbase.sysu.edu.cn/) [35]. We then investigated the TFs that bind to hub genes using CHIPBase (https://rna.sysu.edu.cn/chipbase/) and hTFtarget (http://bioinfo.life.hust.edu.cn/hTFtarget) [36, 37]. The Drug Gene Interaction Database (DGIdb) (https://www.dgidb.org) was used to predict potential pharmacological agents that might interact with the identified hub genes [38]. Finally, we visualized these interaction networks using Cytoscape software. Quantitative real-time polymerase chain reaction (qRT-PCR) Twelve male C57BL/6 mice (6–8 weeks old, body weight 21–25 g), were randomly allocated to two groups: a control group receiving saline, and a fibrosis group receiving intratracheal bleomycin (5 mg/kg). After 21 days, all mice were euthanized for further analysis. The mRNA levels of PRDEGs in mouse lung tissue was examined using qRT-PCR. The primer sequences are listed in Supplementary Table 2. All experimental procedures were approved by the Institutional Review Board and Ethics Committee of the Shanghai Jiao Tong University School of Medicine, Shanghai Ninth People’s Hospital (SH9H-2022-A93-1). Statistical analysis R software (version 4.3.0) was used for data manipulation and analysis. Continuous variables were compared between the two groups using the independent Student’s t-test for normally distributed data and the Mann-Whitney U test for non-normally distributed data. Chi-square or Fisher’s exact tests were used to compare and evaluate categorical variables between the two groups. We also conducted Spearman’s correlation analysis to determine the correlation coefficients among various molecular entities. Unless stated otherwise, all statistical P-values were two-sided, with a significance threshold of p < 0.05 (*p < 0.05, **p < 0.01, and ***p < 0.001). Results Merging and adjusting datasets The basic framework of this study is illustrated in Fig. 1. After standardization, the differences between samples from various sources within the GEO dataset (GSE10667, GSE24206, and GSE53845) were resolved (Supplementary Fig. 1 A, B). We conducted principal component analysis (PCA) on the GEO dataset to evaluate the effectiveness of the adjustment both prior to and following batch effects removal (Supplementary Fig. 1 C, D). The results confirmed that the discrepancies among the samples from different sources were significantly reduced subsequent to batch effect removal. The basic framework of this study Analysis of differentially expressed genes (DEGs) associated with IPF We conducted differential expression analysis within the GEO dataset to explore gene expression differences between the normal and IPF group, identifying 2,655 upregulated genes (logFC > 0, p < 0.05) and 3,215 downregulated genes (logFC < 0, p < 0.05) in the IPF group. A volcano plot was generated to visually depict the outcomes of the differential expression analysis (Fig. 2A). To identify PRDEGs, we intersected all DEGs with known PRGs, resulting in nine PRDEGs: TRADD, SFTA1P, ZBP1, MEFV, PARP1, AIM2, MLKL, CASP6, and MAP3K7. The intersection was shown in a Venn diagram (Fig. 2B). Additionally, we performed a correlation analysis among the PRDEGs and depicted the results using a chord diagram (Fig. 2C). We also performed GSEA to evaluate the impact of gene expression disparities between the normal and IPF group (Fig. 2D). Genes with differential expression were found to be prominently enriched in pathways including Wnt, MAPK, TP53, and Hippo signaling pathways (Fig. 2E-H). Additionally, we conducted GSVA to identify functional enrichment differences and found that gene sets such as the TGF-β and glycolysis pathways exhibited notable differential expression (Fig. 2I). Analysis of differentially expressed genes related to IPF. (A) Volcano plot of the DEGs in the GEO dataset. (B) Venn diagram of DEGs and PRGs in the GEO dataset. (C) Correlation chord diagram of PRDEGs. (D) Gene set enrichment analysis of the GEO dataset. (E-H) DEGs were significantly enriched in Wnt, MAPK, TP53, and Hippo signaling pathways. (I) Gene set variation analysis for the GEO dataset Enrichment analysis of genes in the context of IPF We conducted GO and KEGG enrichment pathway analyses to delineate the biological functions and molecular pathways related to PRDEGs. The findings were visually represented using bubble charts (Supplementary Fig. 2 A, B) and circular network diagrams (Supplementary Fig. 2 C, D). The results of GO and KEGG pathway enrichment analyses on PRDEGs, including logFC values, are illustrated using bar charts (Supplementary Fig. 2E, F). The GO analysis revealed that the nine PRDEGs were significantly associated with pyroptosis, autophagy regulation, execution phase of apoptosis, inflammasome complex, ATAC complex, protein tyrosine kinase binding, death domain binding, and adenosine deaminase activity. The KEGG enrichment analysis highlighted significant pathways such as apoptosis, necroptosis, IL-17, TNF, and MAPK signaling pathways. Expression analysis of PRDEGs The nine PRDEGs exhibited a marked discrepancy in expression levels between the normal and IPF group (Supplementary Fig. 3 A). Notably, TRADD exhibited an extremely significant difference in expression (p < 0.001), while SFTA1P, ZBP1, PARP1, AIM2, and CASP6 also showed high significance (p < 0.01). Furthermore, the expression levels of MEFV, MLKL, and MAP3K7 displayed significant differences (p < 0.05). We also conducted a correlation analysis with the comprehensive expression matrix of the nine PRDEGs, resulting in a correlation heatmap (Supplementary Fig. 3B). We annotated the chromosomal locations of the nine PRDEGs using a circular plot (Supplementary Fig. 3 C). Subsequently, we performed functional similarity analysis of the nine PRDEGs to identify genes associated with cellular, molecular, and biological functions. The analysis revealed that the following eight PRDEGs met the criteria: TRADD, ZBP1, MEFV, PARP1, AIM2, MLKL, CASP6, and MAP3K7 (Supplementary Fig. 3D). Nomogram construction and clinical evaluation of nine PRDEGs We developed a logistic regression model using the expression levels of the nine PRDEGs. The analysis revealed that AIM2 was a significantly more effective predictor in the IPF diagnostic model than the other models. In contrast, PARP1 contributed significantly less to the model than the other predictors (Fig. 3A). Subsequently, we plotted a calibration curve (Fig. 3B), conducted a DCA (Fig. 3C), and illustrated the Receiver Operating Characteristic (ROC) curve for the linear predictors of the logistic regression model (Fig. 3D) to evaluate its performance and clinical utility. The results showed that the diagnostic model exhibited strong predictive performance, achieving a high area under the curve (AUC > 0.9) between the normal and IPF group. In addition, we generated ROC curves for the nine PRDEGs (Fig. 3E-M). TRADD expression moderately distinguished the normal from IPF patients (AUC 0.7–0.9). In contrast, SFTA1P, ZBP1, MEFV, PARP1, AIM2, MLKL, CASP6, and MAP3K7 showed limited accuracy in differentiating the groups (AUC 0.5–0.7). Diagnostic performance of the PRDEGs. (A) Nomogram of the logistic regression model. (B) A calibration curve presented to confirm the reliability of the model. (C) A DCA plot presented to illustrates the clinical utility of the logistic regression model. (D) ROC validation of the logistic regression model. (E-M) Analyses of ROC curves for TRADD, PARP1, CASP6, SFTA1P, ZBP1, AIM2, MEFV, MAP3K7, and MLKL Construction of PANoptosis-related clusters in IPF Two distinct clusters related to PANoptosis were established in IPF by analyzing the expression patterns of nine PRDEGs using consensus clustering methods. As shown, k = 2 was the optimal cluster count for unsupervised classification in the GEO dataset (Fig. 4A). We performed PCA on the IPF sample expression profiles (Fig. 4B) and subsequently present a delta plot of the area under the cumulative distribution function (CDF) curve (Fig. 4C), as well as the corresponding CDF plot (Fig. 4D). This analysis identified two PANoptosis-related clusters in the IPF samples: cluster 1 (high-PANoptosis expression) and cluster 2 (low-PANoptosis expression). Additionally, the comparative analysis of the nine PRDEGs between cluster 1 and 2 revealed that AIM2, MLKL, and ZBP1 exhibited a remarkably high level of statistical significance (p < 0.001), whereas MAP3K7 and PARP1 showed moderate differences (p < 0.01) in their expression between the two clusters (Fig. 4E). The ROC analyses for the nine PRDEGs between clusters (Fig. 4F-J) indicated that AIM2 demonstrated high accuracy (AUC > 0.9), whereas ZBP1, MLKL, MAP3K7, and PARP1 showed moderate accuracy in distinguishing between clusters (AUC 0.7–0.9). Construction of PANoptosis-related clusters. (A) Result of consensus clustering for IPF in the GEO dataset (k = 2). (B) A PCA showing the distinction between the two clusters. (C) The delta plot of the area under the CDF curve. (D) CDF plot of consistency clustering. (E) Comparative expression plots of PRDEGs between the two clusters. (F–J) ROC curves of AIM2, ZBP1, MLKL, MAP3K7, and PARP1 across the two clusters Immune infiltration and functional annotations between the PANoptosis-related clusters The ssGSEA algorithm indicated substantial disparities in the expression profiles of 12 immune cell types between cluster 1 and 2 (Supplementary Fig. 4 A). These immune cells include activated B cells, CD4+, CD8+, and central and effector memory T cells. Additionally, a correlation heatmap showed significant correlations between the immune cells (Supplementary Fig. 4B). Another correlation heatmap displayed the relationships between the nine PRDEGs and the 12 immune cells with significant infiltration levels (Supplementary Fig. 4 C). Subsequently, we performed GSVA to explore the functional alterations among the identified clusters. The KEGG enrichment analysis revealed that cluster 1 exhibited significant upregulation of pathways associated with primary immunodeficiency, B-cell receptor signaling pathway, and intestinal immune network for IGA production. Conversely, cluster 2 showed notable upregulation in pathways related to basal cell carcinoma, histidine metabolism, and adherens junction signaling (Supplementary Fig. 4D). In addition, the GO enrichment analysis showed that cluster 1 was significantly associated with the detection of chemical stimuli, protein-coupled receptor activity, and the neurotransmitter receptor complex. In contrast, cluster 2 was linked to cellular responses to light, UV response, and microtubule anchoring at the microtubule organizing center (Supplementary Fig. 4E). Significant gene module selection We used the WGCNA algorithm to identify significant gene modules by establishing co-expression modules from 50% of the most variable genes in the GEO dataset (Fig. 5A). With a power value of 4, we achieved a network independence of 0.9 and high average connectivity (Fig. 5B). Visualization of gene interconnectivity patterns via Topological Overlap Matrix (TOM) plots and gene clustering trees typically reveals the presence of large co-expression modules. We identified nine distinct gene co-expression modules and visualized them using the TOM plot and gene dendrogram (Fig. 5C, D). In the heatmap plot of topological overlap in the gene network, each row and column represents a single gene: the darker the color, the stronger the co-expression correlation between genes. The block-shaped heatmap regions on the diagonal represent co-expression modules, indicating that these genes exhibit synergistic expression patterns. The color row below the dendrogram indicates module assignments as determined by the Dynamic Tree Cut algorithm. We concurrently analyzed the gene composition of the nine independent modules to evaluate the connection of characteristic genes with clinical traits. The gene expression of the MEgreen module was significantly positively correlated with the IPF phenotype and significantly negatively correlated with the normal phenotype, indicating that it may play a crucial role in the occurrence and progression of IPF (Fig. 5E). The green module, consisting of 1,960 genes, exhibited a strong correlation between gene significance (GS) and module membership (MM), indicating that hub genes within this module are also likely to be highly associated with PANoptosis (Fig. 5F). Construction of the co-expression network of DEGs in IPF. (A) Sample clustering to detect outliers. (B) Confirmation of the soft threshold power. (C) Heatmap plot of topological overlap in the gene network. (D) Gene dendrogram of hierarchical clustering. (E) Association of model genes with clinical traits. (F) Scatterplot of module membership versus gene significance in the green module Similarly, for clusters related to PANoptosis, we used the same soft threshold power and scale-free network coefficient to identify the co-expression modules (Fig. 6A). The dynamic cutting algorithm yielded 12 modules, and we analyzed the interactions between these co-expressed modules (Fig. 6B-D). Correlation analysis demonstrated a robust association of clusters with the magenta module (Fig. 6E), which comprised 207 genes. Subsequently, we generated a scatter plot to depict the correlation between GS and MM within the magenta module (Fig. 6F). Construction of key gene modules associated with PANoptosis-related clusters. (A) Confirmation of the soft threshold power. (B) Gene dendrogram of hierarchical clustering. (C) Clustering of module characteristic genes. (D) Heatmap plot of topological overlap in the gene network. (E) Connection of clinical status with module genes. (F) Scatter plot of module membership versus gene significance in the magenta module Establishment and confirmation of the predictive model We identified 18 hub genes by merging genes associated with GEO dataset and PANoptosis-related clusters in IPF (Fig. 7A). We developed two machine learning models (RF and SVM) to investigate the predictive potential of these genes, achieving high prediction accuracy and a low root mean square error (Fig. 7B, C). Among them, the RF model significantly outperformed the SVM model in terms of residual control, fitting effect, and prediction stability. The top ten variables of both models were displayed in Fig. 7D. Longer blue bars indicate higher gene importance in the RF model, while longer green bars represent greater gene contribution in the SVM model. These genes—identified as high-priority candidates by both models—play a dual critical role in the PANoptosis-IPF association. A significant increase in the root mean square error (RMSE) following permutation testing confirms that these genes are indispensable to this association. The ROC analysis revealed that both the RF and SVM models achieved high AUCs (AUC = 0.908 and AUC = 0.829, respectively) (Fig. 7E). Construction of machine learning models. (A) The intersection of module-associated genes of PANoptosis-related clusters and module-associated genes in IPF. (B) Reverse cumulative distribution of the RF and SVM models. (C) Box plots illustrating the residuals for the RF and SVM models. (D) Key features of the RF and SVM models. (E) ROC curve analyses of the RF and SVM models Using the two machine learning algorithms, we identified six common genes: ASPG, BTNL3, FAM19A2, FOXD2, PLCE1, and SHROOM2 (Fig. 8A). Subsequently, we developed a nomogram model that incorporated these common variables to predict the risk of IPF. The risk was determined by the “total score” derived from this model (Fig. 8B). Evaluation of the model’s prognostic accuracy through calibration curves and DCA indicated high accuracy, aiding clinicians in therapeutic planning (Fig. 8C, D). Additionally, we used binary logistic regression analysis to establish ROC curves for the six common genes (Fig. 8E). The linear predictive expression levels from the logistic regression model exhibited robust accuracy in distinguishing between the normal and IPF group (AUC > 0.8). The predictive model was further validated with the GSE53845 dataset, demonstrating strong discriminative ability (AUC = 0.947) (Fig. 8F). Construction and confirmation of a predictive nomogram. (A) Diagnostic predictive genes selected using the RF and SVM methods. (B) Nomogram to predict the IPF risk using the six common genes (ASPG, BTNL3, FAM19A2, FOXD2, PLCE1, and SHROOM2). (C, D) A calibration curve and DCA plot evaluating the predictive performance of nomogram. (E) ROC curve validation of the logistic regression model. (F) ROC curve validation of the predictive model using the GSE53845 dataset Inflammatory programmed cell death score We calculated the PANoptosis scores for IPF patients based on nine PRDEGs using the ssGSEA algorithm, and divided the IPF patients into high- and low-risk group based on these scores. Subsequently, we visualized the comparative expression of these PRDEGs across different risk groups (Fig. 9A). The analysis indicated that ZBP1, PARP1, AIM2, and MLKL levels were significantly different between the high- and low-risk group (p < 0.001), with MAP3K7 showing moderate differences (p < 0.01). We also generated a co-expression heatmap of the nine PRDEGs (Fig. 9B), and drew ROC curves for the PRDEGs across the two risk groups. The ROC analysis revealed that AIM2 had a high AUC (AUC > 0.9), whereas ZBP1, MLKL, PARP1, and MAP3K7 had moderate AUCs (0.7 < AUC < 0.9), and MEFV had a lower AUC (0.5 < AUC < 0.7) (Fig. 9C-H). Expression of PRDEGs between the high- and low-risk group. (A) Comparison chart of PRDEGs between the two groups. (B) Co-expression heatmap of PRDEGs between the two groups. (C-H) ROC curve analyses for AIM2, ZBP1, MLKL, PARP1, MAP3K7, and MEFV in both groups Immune infiltration analysis in high- and low-risk group We used the CIBERSORT algorithm to assess the associations among the 22 immune cell types in both high- and low-risk group. Immune cell infiltration levels in individual samples were visualized using a bar chart (Fig. 10A). Next, we constructed a box plot contrasting the levels of immune cell infiltration across the two groups. The high-risk group was significantly enriched in memory B cells (p < 0.05), plasma cells (p < 0.01), and activated CD4⁺ T cells (p < 0.001), while the low-risk group was dominated by resting dendritic cells (p < 0.05) (Fig. 10B). These findings provide strong preliminary clues that PANoptosis serves as a marker of humoral immune activation. Furthermore, we generated a heatmap to illustrate the linkage of nine PRDEGs with the degree of immune cell infiltration (Fig. 10C). Another heatmap was generated to emphasize the statistically significant associations between the infiltration of immune cells and their abundances (Fig. 10D). We also used ssGSEA to explore the correlation between the 28 immune cell types across different risk groups and visualized their infiltration abundances using a box plot (Supplementary Fig. 5 A). Additionally, a correlation heatmap was created illustrating statistically significant associations among immune cells across the two groups (Supplementary Fig. 5B). The high-risk group was concurrently enriched with cell-mediated immunity effector cells (activated CD8⁺ T cells, Th1 cells, p < 0.001) and immunosuppressive cells (MDSCs, p < 0.001). We also identified associations between nine PRDEGs and 13 immune cells that exhibited notable infiltration levels (Supplementary Fig. 5 C). Immune infiltration analysis between the high- and low-risk group. (A) Bar chart showing immune cell infiltration in both groups. (B) Comparative chart of immune cell infiltration abundances in both groups. (C) Heatmap illustrating the correlation between PRDEGs and immune cell infiltration abundances. (D) Heatmap showing the correlation analysis results among immune cells Single cell analysis The number of genes, sequencing depth, gene expression complexity, and the proportion of mitochondrial genes detected in each sample are shown (Fig. 11A). After quality control, removal of doublets, and cell cycle processing, PCA analysis was performed for data dimensionality reduction, with the first 30 principal components selected as input for UMAP. UMAP results revealed that the total cells were divided into 24 clusters, which were further classified into 14 cell types using the ScType package (Fig. 11B, C). We investigated the expression of PRDEGs in different cell types and found that most PRDEGs were enriched in immune-related cells (Fig. 11D). Among them, MLKL was predominantly expressed in endothelial cells, non-classical monocytes, and neutrophils, while ZBP, TRADD, PARP1, CASP6 were mainly expressed in memory CD8 + T cells, CD8 + NKT-like cells, γδ-T cells, and natural killer cells. We also focused on the PANoptosis status among different cell types: the UMAP distribution showed the PANoptosis score of each cell (Fig. 11E), the violin plot was used to compare the PANoptosis activity among different cell types (Fig. 11F), and we found that pulmonary alveolar type II cells, memory CD8 + T cells, CD8 + NKT-like cells, and γδ-T cells exhibited higher PANoptosis activity, and most of these cells are immune cells. Single cell analysis of GSE227136. (A) The violin plot of the expression of cells in IPF samples and control samples. (B) UMAP distribution of 24 independent cell clusters. (C) distribution of cell types identified by marker genes. (D) The expression levels of PRDEGs in various cell types. (E) UMAP distribution of PANoptosis scores of cells by using AUCell function. (F) The violin plot of PANoptosis scores of cell types Construction of mRNA-miRNA, mRNA-RNA-RBPs, mRNA-TFs, and mRNA-drugs interaction networks The network of mRNA-miRNA interactions included three hub genes (TRADD, PARP1, MAP3K7) and 50 miRNA molecules (Supplementary Fig. 6 A). The mRNA-RBPs interaction network exhibited six hub genes (AIM2, CASP6, MAP3K7, MLKL, PARP1, and TRADD) and 43 RBPs (Supplementary Fig. 6B). The mRNA-TFs interaction network involved nine hub genes and 114 TFs (Supplementary Fig. 6 C). Furthermore, we pinpointed 35 possible drugs associated with seven hub genes (CASP6, PARP1, MLKL, ZBP1, MAP3K7, MEFV, and TRADD) (Supplementary Fig. 6D). Expression of PRDEGs verified by quantitative real-time polymerase chain reaction (qRT-PCR) To ascertain whether the PRDEGs were pivotal genes related to bleomycin-induced pulmonary fibrosis, we examined the related genes in mouse lung tissue. The outcomes indicated a significant increase in MLKL and AIM2 levels within the Bleomycin group versus the control group (p < 0.05) (Fig. 12). Validation of PRDEGs levels in lung tissue in bleomycin-induced pulmonary fibrosis mouse model. *p < 0.05, **p < 0.01, ***p < 0.001 (normalized to β-actin level) Discussion IPF is a progressive and life-threatening interstitial lung disease with a median survival rate of only 2–3 years [39]. This disease places a significant burden on patients, decreasing quality of life and increasing healthcare costs owing to prolonged hospitalization and the need for advanced treatments [40]. Current treatment options primarily focus on antifibrotic therapies, which have shown limited efficacy, especially in the later stages of the disease [41]. Retrospective studies and surveys have indicated that IPF diagnoses are frequently delayed by 2.1 years [42], highlighting the urgent need for improved diagnostic methods. We analyzed PANoptosis-related predictive models and assessed the immune microenvironment associated with IPF based on scRNA-seq and bulk-seq, potentially revealing new therapeutic targets for the disease. Within the GEO dataset, PRDEGs were expressed at significantly higher levels in patients with IPF compared to healthy controls, suggesting their potential role in disease pathogenesis. Furthermore, KEGG pathway analysis revealed that the PRDEGs were predominantly associated with the TNF and IL-17 signaling pathways, which are crucial for inflammatory immune responses. Subsequently, we analyzed the differences between the two PANoptosis-related clusters based on their immune features, revealing that cluster 1 showed a higher infiltration of immune cells. Furthermore, our GSVA results indicated that pathways linked to Toll-like, NOD-like, and RIG1-like receptor signaling pathways were markedly upregulated in cluster 1, which was associated with a worse prognosis. This suggests that inflammation within the lung tissues of individuals might contribute to the pathogenesis of IPF, supported by prior studies that emphasized the role of inflammation in IPF progression [43, 44]. In contrast to conventional prediction models, various machine learning approaches have demonstrated enhanced robustness and reduced error rates in predicting disease incidence [45]. As revealed by the residual distribution and residual squared boxplots, the ensemble learning characteristic of RF effectively handles the high-dimensional noise in gene expression data and is more sensitive to the “heterogeneity of PANoptosis gene expression” (consistent with the molecular heterogeneity of IPF), thus providing greater stability in small-sample, multi-feature IPF diagnosis. Furthermore, compared with the SVM model, RF achieved superior diagnostic accuracy (AUC = 0.908) and greater potential for clinical application. Leveraging the favorable performance of these two algorithms, we implemented RF and SVM to identify six genes with significant predictive importance: ASPG, BTNL3, FAM19A2, FOXD2, PLCE1, and SHROOM2. These genes play a “dual critical role” in the PANoptosis-IPF association – they are involved in PANoptosis regulation or the IPF pathological process and are also stable in distinguishing IPF patients from normal controls. Existing literature has linked these six genes to multiple biological processes and pathways—for instance, FAM19A2 (also known as TAFA-2) has an intronic variant associated with FEV1/FVC; FOXD2-related long noncoding RNA (FOXD2-AS1) is a cancer-related lncRNA correlated with poor prognosis in non-small cell lung cancer; PLCE1 can alleviate lipopolysaccharide-induced acute lung injury; and SHROOM2, as a RhoA-ROCK pathway mediator, inhibits tumor metastasis in nasopharyngeal carcinoma [46–49]. While these genes have been implicated in key pathological processes of IPF (e.g., lung epithelial cell protection, immune regulation, and cell proliferation regulation), their specific roles in IPF progression remain underexplored. Among them, ASPG, which is involved in the regulation of cell proliferation and associated with the RAS pathway [50], might contribute to the aberrant repair of alveolar epithelial cell injury, thereby acting as a key driver of IPF pathogenesis. As an immune regulatory molecule, BTNL3 ranked highly in both RF and SVM analyses, with its expression showing a significant positive correlation with γδ-T cell infiltration and IL-17 pathway activation in IPF samples. Building on previous findings [51], we hypothesize that BTNL3 might promote the activation of pulmonary fibroblasts by activating PANoptosis in immune cells and triggering the release of inflammatory factors (e.g., IL-17, TNF), ultimately leading to pulmonary fibrosis. This hypothesis aligns with the KEGG enrichment results for the “IL-17 signaling pathway” and is consistent with the “enrichment of immune cells in the high-risk group” observed in the immune infiltration analysis. In summary, ASPG and BTNL3 bridge PANoptosis with the alveolar repair mechanisms and/or immune microenvironment in IPF, highlighting their potential as molecular targets for subsequent functional validation and targeted therapeutic interventions. We constructed a nomogram using these six genes and validated it using calibration plots and ROC curves. These findings demonstrate that the nomogram exhibits notable reliability and consistency, and particularly shows significant clinical diagnostic performance in the independent dataset GSE53845 (AUC = 0.947). This indicates that the diagnostic performance of the six-gene nomogram model is significantly superior to that of single biomarkers, fully reflecting the reliability and practical value of multi-gene combined molecular diagnosis in the clinical detection of IPF. This model compensates for the limitations of conventional imaging and pulmonary function tests, can be used for the early auxiliary diagnosis of patients with “suspected IPF but atypical imaging manifestations”, and has promising potential for clinical translational application. Although there have been sporadic studies on PANoptosis in IPF, our study first proposed the scoring concept of “inflammatory programmed cell death score”, namely PANoptosis score. Based on nine PRDEGs, the activity of PANoptosis in IPF patients was quantified using relevant algorithms, which filled the gap in the lack of quantitative evaluation tools in this field. In addition, we established a quantitative link between the PANoptosis score and immune phenotypes, which offered support for strategies targeting the immune microenvironment. Consistent with previous findings [52–55], immune infiltration analysis across different risk groups showed a significant increase in NKTs, Tfh, and type 17 T helper cells in the high-risk group. Moreover, in comparison to the low-risk group, the high-risk group exhibited increased levels of certain immune cells, including activated B cells, CD4 + T cells, CD8 + T cells, central memory CD4 + T cells, effector memory CD4 + and CD8 + T cells, immature B cells, myeloid-derived suppressor cells, and memory B cells. At the single-cell level, as a cutting-edge technology that reveals transcriptome alterations at the single-cell level, scRNA-seq has unique advantages in exploring the heterogeneity of cell populations and identifying cell types associated with IPF. After annotation, we successfully divided 24 cell clusters into 14 cell types. In this study, our primary focus was to compare the PANoptosis status of cell populations in IPF tissues. We scored each cell based on the expression of nine PRDEGs, and the results revealed that most PRDEGs (especially ZBP1, TRADD, PARP1, CASP6) are enriched in immune cell types such as memory CD8 + T cells, CD8 + NKT-like cells, and γδ-T cells, and these immune cells exhibit high PANoptosis activity. The results were in agreement with those from the transcriptome-level immune infiltration analysis. PANoptosis is an inflammatory and lytic cell death pathway governed by PANoptosome complexes, which distinguishes it from other forms of cell death. To date, various PANoptosome complexes, including ZBP1, AIM2, RIPK1, and NLRP12 PANoptosomes, have been characterized in detail [56]. Our study demonstrated a considerable elevation in the expression of MLKL and AIM2 within the lung tissue of bleomycin-induced pulmonary fibrosis model, supporting the results of previous studies [56]. MLKL, a critical marker of necroptosis, has a key impact on downstream necrosome formation. The activation of necroptosis and subsequent formation of pores by MLKL can induce pyroptosis by damaging cellular membranes [57]. AIM2, a biomarker of pyroptosis, forms a complex with several proteins, including pyrin, ZBP1, ASC, caspase-1, caspase-8, FADD, RIPK1, and RIPK3, leading to formation of the AIM2 PANoptosome, which promotes PANoptosis [56]. Although ZBP1 was identified as the primary sensor of the ZBP1-PANoptosome, we found no significant changes in ZBP1 levels. After all, this study mainly reflects changes in gene transcription levels based on transcriptomic data and qRT-PCR experiments, and cannot rule out the direct impact of post-translational modifications (PTMs) on protein expression and function in the absence of significant transcriptional changes. There are several limitations in our study. Firstly, these datasets in this study lacked essential clinical information, such as pulmonary function metrics (forced vital capacity, FVC; diffusing capacity of the lung for carbon monoxide, DLco), which limited the clinical translational value of our prognostic model. Second, Protein-level validation experiments, such as western blot and immunohistochemistry (IHC), should be performed in subsequent studies. Finally, exploring how PRGs regulate PANoptosis and drive IPF progression—including via PTMs of key pathway molecules—is critical to deciphering IPF’s underlying molecular processes. Conclusions In conclusion, our research examined the potential function of PANoptosis in IPF-related processes. Beyond that, we initially proposed the scoring concept referred to as “inflammatory programmed cell death score” and distinguished two risk groups of IPF. Moreover, the associations between PANoptosis activity and cellular immunity were explored via integrating scRNA-seq and bulk RNA-seq data. It should be noted that further studies are required to elaborate on the biological roles and mechanisms of these genes in driving the pathogenesis of IPF. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Author contributions ZW conducted the data analysis and wrote the manuscript. QL helped collect and analyze the partial data. YC, LG, LY, JL and YX helped revise the manuscript. TZ conducted the animal experiments and offered funding support. XG reviewed the manuscript and supervised the research. All authors approved the final manuscript. Funding This study was supported by the National Natural Science Foundation of China (grant 82000041) and the Shanghai Municipal Health Commission Health Industry Research Special Young Project (grant 20234Y0091). Data availability The raw data utilized for this research are available through downloading datasets from [https://www.ncbi.nlm.nih.gov/geo/](https:/www.ncbi.nlm.nih.gov/geo), including GSE10667, GSE24206, GSE53845 and GSE227136. The original contributions presented in the study are included in the article/supplementary material. Additional inquiries should be made to the corresponding author. Declarations Competing interests The authors declare no competing interests. Institutional review board statement The study involving animal experiments was approved by the Institutional Review Board and Ethics Committee of the Shanghai Jiao Tong University School of Medicine, Shanghai Ninth People’s Hospital (SH9H-2022-A93-1). Informed consent Not applicable. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Zhihua Wang and Quanlei Li contributed equally to this work. References Associated Data Supplementary Materials Supplementary Material 1 Data Availability Statement The raw data utilized for this research are available through downloading datasets from [https://www.ncbi.nlm.nih.gov/geo/](https:/www.ncbi.nlm.nih.gov/geo), including GSE10667, GSE24206, GSE53845 and GSE227136. The original contributions presented in the study are included in the article/supplementary material. Additional inquiries should be made to the corresponding author.