RAMP: response-aware multi-task learning with contrastive regularization for cancer drug response prediction

Area of Science:

• Computational biology
• Bioinformatics
• Genomics

Background:

• Precision cancer medicine relies on accurate prediction of drug sensitivity from patient multiomics profiles.
• Existing models face challenges due to complex feature interactions and imbalanced drug response data in public databases.

Purpose of the Study:

• To develop a novel framework, Response-Aware Multi-task Prediction (RAMP), for accurate multidrug response prediction.
• To overcome limitations of existing methods in handling complex features and imbalanced datasets.

Main Methods:

• Utilized a Bayesian neural network integrated with soft-supervised contrastive regularization.
• Employed response-aware negative sampling for network embedding, incorporating cell line-drug response information.
• Developed a comprehensive approach for selecting and utilizing drug response features to address data imbalance.

Main Results:

• Achieved an Area Under the Receiver Operating Characteristic Curve (AUC-ROC) > 89%, Area Under the Precision-Recall Curve (AUC-PR) > 59%, and F1-score > 52% on the Genomics of Drug Sensitivity in Cancer dataset.
• Outperformed existing methods on both balanced and imbalanced datasets.
• Successfully predicted numerous missing drug responses not present in public databases.

Conclusions:

• RAMP demonstrates suitability for high-throughput prediction of cancer drug sensitivity.
• The framework can effectively guide cancer drug selection processes in clinical settings.
• RAMP offers a robust solution for improving the accuracy and utility of multiomics data in precision oncology.