Identification of potential biomarkers and therapeutic targets for liver cirrhosis based on Mendelian randomization and machine learning.

BACKGROUND

Liver cirrhosis(LC) represents the end stage of chronic liver disease, yet reliable molecular markers remain limited. This study aimed to uncover potential diagnostic biomarkers and therapeutic targets for LC.

METHODS

We integrated differential gene analysis from LC datasets in the Gene Expression Omnibus (GEO) with Mendelian randomization (MR) using eQTLGen and FinnGen summary data to prioritize LC-associated genes. LASSO, SVM-RFE, RF, and XGBoost algorithms were applied to refine candidate genes. Based on these, we constructed a nomogram risk prediction model and evaluated by receiver operating characteristic (ROC) curve. Gene set enrichment analysis (GSEA) and immune infiltration profiling were conducted to explore potential biological functions. Additionally, potential therapeutic compounds targeting these genes were screened using Drug Signatures Database (DSigDB) via Enrichr platform. Finally, hub genes were validated by immunohistochemistry (IHC).

RESULTS

Through integrative analysis,we identified five hub genes: ENPP2, FAM134B, PPARGC1A, SLFN11, and TRIM22. A nomogram based on these genes demonstrated strong predictive performance (AUC = 0.944 in the training set; AUC = 0.909 in the validation set). GSEA linked these genes to antigen processing, cell adhesion, and immune regulation. Immune infiltration analysis indicated that abnormal levels of resting NK cells (P < 0.001), M2 macrophages (P = 0.002), activated dendritic cells (P < 0.001), and neutrophils (P = 0.038) in LC. Drug prediction provides promising treatment options for LC, including valproic acid and tamoxifen.

CONCLUSION

Our integrated approach identified five hub genes associated with LC, providing valuable clues to predict and treat LC.