Development of a prognostic prediction signature for idiopathic pulmonary fibrosis by integrating multiple programmed cell death-related genes and machine learning algorithms.

Background: Idiopathic pulmonary fibrosis (IPF) is a chronic progressive lung disease with a poor prognosis, severely limits patient survival. Various forms of programmed cell death (PCD) play significant roles in the progression of IPF. However, research on the comprehensive effects of different PCD patterns on the prognosis of IPF is still insufficient. This study aims to integrate multiple PCD-related genes to construct a prognostic model for IPF, providing new insights for individualized clinical prognosis assessment and facilitating more precise therapeutic strategies. Methods: This study used multiple machine learning analysis methods, multiple datasets, and integrated 16 types of PCD related genes for prognostic signature construction in IPF. The infiltration of different immune cell types in IPF samples was also assessed. The model's predictive performance was evaluated through validation with external datasets. Finally, on the basis of the selected key genes, potential drugs for IPF were identified using the Broad Institute's Connectivity Map databases. Results: Differential analysis of five Gene Expression Omnibus (GEO) datasets identified 1,491 PCD-related hub genes. Integration with overall survival (OS) data revealed 241 differentially expressed genes (DEGs) prognostic for IPF. Key PCD types included apoptosis, immunogenic cell death, necrosis, necroptosis, and ferroptosis. Functional enrichment highlighted DEG involvement in cell death/apoptosis and the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways like apoptosis, tumor necrosis factor (TNF), and NF-κB signaling. Using machine learning and least absolute shrinkage and selection operator (LASSO) regression, seven prognostic biomarkers (TLR2, PDCD1LG2, AKT3, SLC1A4, ANTXR2, HTRA1, and TIMP1) were identified. The derived cell death score (CDS) effectively stratified high- and low-risk patients in training (GSE70866) and validation (GSE70867) cohorts [the area under the curve (AUC) >0.75]. A CDS-based nomogram demonstrated high prognostic accuracy (max AUC =0.907). Immune analysis revealed B cell enrichment in IPF and significant monocyte/activated natural killer (NK) cell infiltration in high-risk patients, correlating with worse OS. Screening identified 25 potential therapeutic small molecules [e.g., acetylcholinesterase inhibitors and histone deacetylase (HDAC)]. Quantitative real-time polymerase chain reaction (qRT-PCR) validation in a bleomycin-induced mouse model confirmed differential expression of the seven model genes, aligning with transcriptomic predictions. Conclusions: The prognostic signature based on PCD-related genes provides new biomarkers for the prognostic assessment of IPF, and has high predictive accuracy. Additionally, the identified potential drugs offer new directions for the treatment of IPF, laying the foundation for future individualized therapies.