Integrated single-cell and transcriptomic profiling identifies machine-learning-based pyroptosis biomarkers in IBD.

Background: Pyroptosis, an inflammatory form of programmed cell death, contributes to intestinal inflammation in inflammatory bowel disease (IBD), but the key cell types and regulatory genes remain unclear. Methods: We analyzed single-cell RNA-seq data from intestinal mucosa to assess pyroptosis-related gene expression signatures using AUCell, AddModuleScore, and GSVA. High- and low-pyroptosis cells were compared to identify core genes. Findings were validated in bulk peripheral blood transcriptomic data. Machine learning (LASSO, GBM, SVM, Boruta, Random Forest) identified optimal diagnostic genes, and functional enrichment explored biological roles. Results: Pyroptosis was most active in macrophages, epithelial cells, and neutrophils. Six cell subsets consistently exhibited high pyroptosis. Nineteen core pyroptosis-related genes were identified, 16 of which were upregulated in both mucosal and blood samples. Functional analysis linked these genes to apoptosis and reactive oxygen species pathways. Machine learning highlighted six key diagnostic genes-BASP1, LITAF, NAMPT, PHACTR1, PLAUR, PPIF-with BASP1 showing the strongest performance (AUC = 0.935). Conclusions: Pyroptosis is highly active in specific immune and epithelial cells in IBD. Six identified genes show potential as non-invasive diagnostic biomarkers, offering insights into disease mechanisms and therapeutic targets.