Integrative Machine Learning Analysis of Programmed Cell Death Pathways Identifies Novel Diagnostic Biomarkers for...

Atrial fibrillation (AF) is a leading cause of stroke, heart failure, and mortality, yet the molecular mechanisms remain incompletely defined.

We integrated bulk transcriptomes from GEO with weighted gene co-expression network analysis, consensus clustering, and a 12-algorithm machine-learning pipeline (66 model combinations) to map programmed cell death (PCD) pathways and pinpoint diagnostic genes. Immune infiltration was profiled by CIBERSORT, xCell, and ssGSEA. Hub-gene expression was validated in an HL-1 atrial pacing model and in peripheral blood mononuclear cells (PBMCs) from patients with persistent AF.

Four hub genes-SGPL1, NPC2, PTGDS, and RCAN1-were identified and incorporated into a nomogram and a PCD-based risk score (PCDscore). The nomogram showed robust discrimination in the training cohort and two independent validation datasets. Patients with a high PCDscore exhibited markedly increased immune-cell infiltration and dysregulated immune modulators, with macrophages consistently enriched across algorithms. qRT-PCR confirmed up-regulation of SGPL1, NPC2, and RCAN1 and down-regulation of PTGDS in AF cell models; NPC2 and SGPL1 were further elevated in PBMCs from AF patients.

Our integrative framework reveals PCD-linked remodeling in AF and nominates SGPL1, NPC2, PTGDS, and RCAN1 as candidate diagnostic biomarkers, providing a PCD-based nomogram and risk score that may inform patient stratification and hypothesis-generating targeted interventions.