Omics Integration via Explainable Deep Learning to Decipher Gene-Environment Interactions in the Pathogenesis and Progression of Chronic Fatty Liver Disease

• Billy Elly

  Lautech

  Author

Chronic fatty liver disease, encompassing non-alcoholic fatty liver disease (NAFLD) and its progressive form metabolic dysfunction-associated steatohepatitis (MASH), represents a growing global health burden affecting approximately 38% of adults worldwide . The pathogenesis involves complex interplay between genetic susceptibility, environmental exposures, and metabolic disturbances that single-omics approaches cannot adequately capture. Current predictive models lack interpretability, limiting clinical translation and mechanistic understanding. This study addresses these gaps by developing an explainable deep learning framework integrating multi-omics data (genomics, transcriptomics, proteomics, metabolomics) with environmental exposure profiles to model gene-environment interactions in chronic fatty liver disease progression. Using a retrospective cohort of 1,247 patients with longitudinal multi-omics and exposure data, we implemented a multi-branch attention-based neural network with SHAP (SHapley Additive exPlanations) for model interpretability. The integrated framework achieved 89.4% accuracy (AUC = 0.92) in predicting progression from steatosis to MASH and fibrosis, significantly outperforming single-omics models (p < 0.001). Key predictors included PNPLA3 genotype interactions with air pollution exposure (Oxwt), identified through SHAP analysis , and lipid metabolism biomarkers (LPA, LCAT, CD5L) . The framework successfully stratified patients into risk groups and identified interpretable biomarker panels. This research provides a replicable, interpretable computational framework for precision risk stratification and biomarker discovery, with implications for early intervention strategies and personalized monitoring in chronic liver disease.

• pdf

Copyright (c) 2026 Billy Elly (Author)

This work is licensed under a Creative Commons Attribution 4.0 International License.