Randomized trial demonstrates improved drug synergy prediction in cell lines, indicating robust generalization ability.
Key Points
The research aims to enhance the prediction of drug synergy by addressing limitations in existing machine learning models regarding cell line specificity and generalization.
Proposed a domain generalization-driven framework called PDTSyn with disentangled representation learning.
Focused on generating cell line-adaptive attention parameters through a parameter-decomposed transformer.
Implemented a dual regularization strategy using Kullback-Leibler divergence and cell-line discriminative loss.
PDTSyn outperformed state-of-the-art baselines in standard evaluations across various datasets.
Consistent performance in challenging settings such as unseen cell lines, drugs, and drug pairs, demonstrating robustness to distribution shifts.