Tissue-Anchored Graph Neural Network Framework for Non-Invasive Detection of Non-Alcoholic Steatohepatitis

Overview

Non-Alcoholic Steatohepatitis (NASH) is a severe, chronic form of Non-Alcoholic Fatty Liver Disease (NAFLD). An estimated 16 to 17 million Americans have NASH, and another 80 to 100 million individuals have NAFLD and are at risk for developing NASH. Medical intervention is necessary for people with NASH to prevent progression to liver failure. However, the current diagnostic method for NASH still relies on an invasive liver biopsy. In this project, the goal was to identify biomarkers of NASH using a tissue-anchored graph neural network. The output of this work was intended to facilitate a tissue-anchored blood biomarker panel that can detect NASH with high accuracy while preserving biological relevance. To accomplish this, NASH associated RNA sequence data were collected from Gene Expression Omnibus, and additional proteomics and blood-based data were gathered from other databases. After preprocessing, liver gene networks were constructed to model key biological processes involved in disease progression. The graph neural network leveraged co-expression structure to prioritize candidate biomarkers, which were then evaluated through pathway enrichment, logistic regression panel testing, STRINGdb interaction analysis, blood-based validation, and mechanistic PhysiCell simulation. The model demonstrated strong cross-dataset generalization when trained on GSE126848 and evaluated on GSE239422, achieving an AUROC of 0.862, AUPRC of 0.892, F1 score of 0.700, and accuracy of 0.728. Prioritized genes clustered in pathways related to lipid metabolism, inflammation, detoxification, bile acid dysregulation, and metabolic stress, and several candidates, including CCAR1, PRPF40A, KPNA2, and LGALS3, also showed evidence of detectability in blood.

Competition history

• CSEF 2026 Computational Science (Senior Division) · Entry S-07-43

Related projects

Closest projects by meaning, across every fair and year in the corpus.

Browse more like this

• All Computational Sciences projects
• CSEF Computational Sciences projects
• Every CSEF 2026 project