Candir, E. B.; Kuru, H. I.; Rattray, M.; Cicek, A. E.; Tastan, O.

Combinatorial drug therapy holds great promise for tackling complex diseases, but the vast number of possible drug combinations makes exhaustive experimental testing infeasible. Computational models have been developed to guide experimental screens by assigning synergy scores to drug pair-cell line combinations, where they take input structural and chemical information on drugs and molecular features of cell lines. The premise of these models is that they leverage this biological and chemical information to predict synergy measurements. In this study, we demonstrate that replacing drug and cell line representations with simple one-hot encodings results in comparable or even slightly improved performance across diverse published drug combination models. This unexpected finding indicates that current models take a \'shortcut\'; rather than learning the intended chemical or biological information, they exploit these representations as mere identifiers and capture general co-variation patterns of synergy measurements. While the current synergy prediction models can still aid in prioritizing drug pairs within a panel of tested drug and cell lines, our results highlight the need for better strategies to learn from intended features and generalize to unseen drugs and cell lines.