Pisces: A multi-modal data augmentation approach for drug combination synergy prediction.

Abstract

Drug combination therapy is promising for cancer treatment by reducing resistance and improving efficacy. Machine learning approaches to predicting drug combinations require massive training data. Here, we propose Pisces, a novel machine learning approach for drug combination synergy prediction. The key idea is to augment the sparse dataset by creating multiple views for each drug combination based on different modalities. We combined eight modalities of a drug to create 64 augmented views. By treating each augmented view as a separate instance, Pisces can process any number of drug modalities, circumventing the issue of missing modality. Pisces obtained state-of-the-art results on cell-line-based and xenograft-based drug synergy predictions and drug-drug interaction prediction. By interpreting Pisces's predictions using a genetic interaction network, we identified a breast cancer drug-sensitive pathway from BRCA cell lines. Collectively, the results show that Pisces effectively predicts drug synergy and drug-drug interactions through data augmentation and can be applied to various biological applications.