Multi-omics nominates VDAC2 as a candidate protective locus in sepsis-associated cholesterol dysregulation.

Sepsis, a life-threatening condition, involves dysregulated cholesterol metabolism critical for immune regulation and cellular processes. This study employed multi-omics and machine learning to explore cholesterol metabolism in sepsis, aiming to identify novel therapeutic targets. Transcriptome and single-cell RNA sequencing data for sepsis were retrieved from the Gene Expression Omnibus (GEO) database. The limma package and WGCNA co-expression network were used to screen genes, hybridized with cholesterol metabolism genes (CMGs) to identify hub genes. Machine learning algorithms screened pivotal genes to construct diagnostic model, validating performance via multi-cohort Receiver Operating Characteristic (ROC) curve. Non-negative matrix factorization (NMF) based molecular typing using CMGs, and integration of 101 machine learning algorithms built prognostic models. Single-cell analysis characterized expression patterns of pivotal genes and key subsets. Causal effects and phenotypic associations of target genes were evaluated using Summary data-based Mendelian Randomization (SMR) and PheWAS. Integrated transcriptomic analysis identified three key genes (VDAC1, VDAC2, and LDLRAP1) associated with dysregulated cholesterol metabolism in sepsis. Machine learning-based diagnostic models exhibited high predictive accuracy. NMF clustering revealed two molecular subtypes, with Cluster 1 characterized by immunosuppression and metabolic reprogramming, linked to poorer prognosis. A machine learning model integrating 101 algorithms predicted 28-day mortality. The single-cell transcriptome atlas identified CD14