Shteto, A., Boulahfa, J., Etcheto, A. et al. Genomic biomarkers of liver toxicity risk from UK Biobank data. Pharmacogenomics J 25, 30 (2025).
Toxic liver disease represents one of the most serious safety challenges in drug development and clinical practice. In this new study, Ariana Pharmaceuticals harnessed the power of explainable Artificial Intelligence (eXp AI) through its proprietary KEM® (Knowledge Extraction & Management) platform to uncover genomic biomarkers associated with increased mortality risk in liver toxicity.
Using real-world genomic and clinical data from the UK Biobank, Ariana’s researchers analyzed 225 patients diagnosed with toxic liver disease (ICD-10 K71). Through an integrative approach combining 36,394 liver-related SNPs from key biological pathways—such as KEGG Non-Alcoholic Fatty Liver Disease (NAFLD) and Adverse Outcome Pathways (AOP) for liver injury—the study identified 15 significant genetic variants linked to heightened mortality.
Among these, the PRKAG2 rs73158145 variant was shown to reduce median survival by 3.5 times among homozygous carriers. The AI-selected SNP model achieved 85% mean accuracy, outperforming conventional machine learning models (68.9%), with high sensitivity (66.2%) and specificity (91.8%).
These results mark a major advance in the discovery of pharmacogenomic safety biomarkers, offering interpretable and clinically meaningful insights that could improve patient safety, risk stratification, and regulatory decision-making.