Comparative Analysis of Drug Synergy Prediction Models

This article comprehensively reviews computational models for predicting drug interactions in cancer therapy, focusing on ma- chine learning and deep learning. Synergism refers to the combination of two drugs having more than equal e ff ects than their in- dividual e ff ects, providing significant benefits such as increased performance, reduced drug resistance, and reduced toxicity. How- ever, traditional clinical trials to discover drug combinations are expensive and time-consuming, requiring computational al- ternatives. This article compares three modeling tools (DeepSynergy, TranSynergy, and MatchMaker) using di ff erent mod- els and prediction targets. While DeepSynergy and TranSynergy predict synergistic scores, TranSynergy outperforms DeepSyn- ergy in accuracy across multiple tissue types, primarily through the synergistic combination of self-monitoring and genes. Match- Maker, which predicts Loewe joint scores, outperformed both models in terms of accuracy, with lower mean square er- rors and higher Pearson and Spearman correlation values. We also explore tissue-specific performance, particularly issues aris- ing from poor cell lines, and discuss the implications of general modeling—the importance of trialing to accelerate drug combina- tion and inform oncology reality.