Integrating Genome-Scale Metabolic Modeling with Machine Learning Improves Gene Essentiality Prediction in Triple-Negative Breast Cancer

Triple-negative breast cancer (TNBC) poses a significant therapeutic challenge owing to its aggressiveness and limited treatment options. Here, we integrated genome-scale metabolic modeling with machine learning to improve gene essentiality prediction and identify candidate therapeutic targets for TNBC. Cell-line-specific genome-scale metabolic models were reconstructed for 50 breast cancer cell lines using RNA-sequencing from Cancer Dependency Map (DepMap). Metabolic reaction flux distributions derived from minimization of metabolic adjustment (MOMA) were used as features to train a random forest classifier, with DepMap gene dependency scores as ground truth labels. This integrative approach outperformed the MOMA alone for gene essentiality prediction, increasing sensitivity from 0.37 to 0.55. The model identified 57 TNBC-specific essential genes, including Enolase 1 (ENO1), that were missed by MOMA-based prediction. Furthermore, 30 synthetic lethal partners of succinate dehydrogenase subunit A (SDHA) were predicted in TNBC cell lines. This framework demonstrates the utility of combining metabolic modeling with machine learning for identifying context-specific cancer vulnerabilities.