Dual-view jointly learning improves personalized drug synergy prediction.

Area of Science:

• Computational biology
• Pharmacogenomics
• Bioinformatics

Background:

• Accurate estimation of synergistic drug combinations is crucial for precision medicine.
• Existing computational methods often lack reliability, particularly for cross-dataset predictions, due to complex drug interactions and cancer sample heterogeneity.

Purpose of the Study:

• To develop a robust computational method for predicting sample-specific drug combination effects.
• To improve the accuracy and generalization capabilities of drug synergy predictions.

Main Methods:

• Proposed JointSyn, a dual-view jointly learning framework.
• Utilized drug and cell features for predicting sample-specific drug combination effects.
• Evaluated performance across various benchmarks and explored generalization with fine-tuning.

Main Results:

• JointSyn significantly outperforms existing state-of-the-art methods in predictive accuracy and robustness.
• Each view in JointSyn captures complementary drug synergy-related characteristics.
• Fine-tuned JointSyn demonstrates improved generalization for novel drug combinations and cancer samples.
• Generated a pan-cancer drug synergy atlas, revealing differential patterns among cancers.

Conclusions:

• JointSyn offers a powerful tool for predicting drug synergy, supporting personalized combinatorial therapies.
• The method enhances the development of targeted cancer treatments by providing reliable synergy estimations.