An Exploratory Transcriptomic Classification Model for Psoriasis Based on Apoptosis-Associated and Proliferation–Apoptosis-Coupled Genes Using Explainable Machine Learning

This study aimed to integrate apoptosis-associated and proliferation–apoptosis-coupled transcriptomic signatures with explainable machine learning to construct an exploratory molecular classification model for psoriasis. Transcriptomic datasets GSE30999 and GSE53552 were merged as the skin-tissue training cohort, and GSE55201, a whole-blood transcriptomic dataset, was used as an independent cross-tissue external validation cohort. Differential expression analysis identified 3707 DEGs, and intersection with GeneCards apoptosis-related genes yielded 894 overlapping genes. After PPI-based hub gene selection, eight machine learning algorithms were exploratorily compared within a preselected 25-gene feature space. DALEX-based permutation feature importance analysis identified a five-gene apoptosis-associated and proliferation–apoptosis-coupled signature comprising CCNB1, KIF11, HDAC1, TPX2, and MELK. The five-gene model achieved an AUC of 0.966 in the training cohort and 0.811 in the external whole-blood validation cohort, indicating moderate cross-tissue generalizability. Calibration and decision-curve analyses were performed only in the training cohort and should be interpreted as exploratory analyses rather than evidence of clinical utility. Overall, this study provides an interpretable transcriptomic classification framework for distinguishing psoriasis from healthy controls, while its ability to differentiate psoriasis from clinically similar dermatoses remains to be validated in independent disease-control cohorts.