Abstract
The present study utilized large-scale genome-wide association studies (GWAS) summary data (731 immune cell subtypes and three primary sclerosing cholangitis (PSC) GWAS datasets), meta-analysis, and two PSC transcriptome data to elucidate the pivotal role of Tregs proportion imbalance in the occurrence of PSC. Then, we employed weighted gene co-expression network analysis (WGCNA), differential analysis, and 107 combinations of 12 machine-learning algorithms to construct and validate an artificial intelligence-derived diagnostic model (Tregs classifier) according to the average area under curve (AUC) (0.959) in two cohorts. Quantitative real-time polymerase chain reaction (qRT-PCR) verified that compared to control, Akap10, Basp1, Dennd3, Plxnc1, and Tmco3 were significantly up-regulated in the PSC mice model yet the expression level of Klf13, and Scap was significantly lower. Furthermore, immune cell infiltration and functional enrichment analysis revealed significant associations of the hub Tregs-related gene with M2 macrophage, neutrophils, megakaryocyte-erythroid progenitor (MEP), natural killer T cell (NKT), and enrichment scores of the autophagic cell death, complement and coagulation cascades, metabolic disturbance, Fc gamma R-mediated phagocytosis, mitochondrial dysfunction, potentially mediating PSC onset. XGBoost algorithm and SHapley Additive exPlanations (SHAP) identified AKAP10 and KLF13 as optimal genes, which may be an important target for PSC.