Predicting DNA damage response using synthetic cell painting profiles and experimental analysis

Detecting DNA damage response (DDR) using cell painting profiles is challenging due to limited sample sizes and skewed class distributions. We established a robust classification framework to enhance DDR prediction based on synthetic data. Using the idr-0080 dataset, we generated synthetic profiles with the Gaussian copula, CTGAN, VAE, and CopulaGAN algorithms, and assessed their quality through fidelity metrics. Among four classifiers evaluated with real and/or synthetic data under preserved or resolved class imbalance, an SVM trained on real data augmented by Gaussian copula-generated synthetic data achieved the best performance (F1-score = 0.87, AUROC = 0.94). SHAP analysis highlighted key predictive morphological features. The model successfully identified known and previously unreported DDR inducers in the external cpg-0012 dataset, which were experimentally confirmed by γH2AX marker accumulation and reduced cell viability. Overall, our machine learning framework integrating synthetic cell painting profiles effectively predicted DDR, providing a scalable virtual prescreening approach for drug discovery.