Integrative Multi-Omics Identifies a Causal Immune Gene Panel for Sepsis Diagnosis and Targeted Therapy

Sepsis diagnosis remains challenging due to the lack of specific biomarkers that reflect immune dysregulation, as conventional markers lack immune specificity and causal grounding.

We performed a two-sample Mendelian randomization meta-analysis to identify immune phenotypes causally linked to sepsis, with associated genes mapped via multi- -omics annotation. Integrating three GEO datasets (n = 1,039), we constructed a diagnostic panel using six machine learning models, validated its cell-type specificity via singlecell RNA-seq, confirmed protein-level expression using public proteomic datasets, and predicted potential therapeutic compounds targeting the panel genes.

We characterized protective and risk immune phenotypes in sepsis, identifying 119 genes enriched in immune-metabolic pathways. Machine learning refined these to a six-gene panel (LY9, CD244, OLFML2B, TMEM59, TBC1D2, and FCER1G), with NNET achieving an AUC of 0.973 (95% CI: 0.951-0.996). Single-cell transcriptomics revealed cell-type-restricted expression, while proteomic data corroborated differential expression of key genes in sepsis-related contexts. Drug repurposing analysis prioritized candidates, highlighting its translational potential.

This six-gene panel links immune-based diagnosis directly to therapeutic intervention. The identified genes not only serve as diagnostic biomarkers but also point to repurposable compounds. By connecting causal immune genes to existing drugs, the panel provides a clinical pathway from biomarker detection to targeted sepsis therapy.

This study establishes a causally grounded, diagnostically robust six-gene immune panel for sepsis with demonstrated therapeutic potential, offering a translational framework for biomarker-guided diagnosis and targeted immunotherapy.