A universal model for drug-receptor interactions

The genomic landscape of disease holds, in principle, the information required for rational therapeutic design. Genes encode proteins whose functions are tightly coupled to their three-dimensional structures via non-bonded interactions. Since the late 1970s, the advent of macromolecular crystallography inspired the notion that structural knowledge alone could enable a “lock-and-key” approach to drug design. However, this framework has failed to catalyze a step-change in the generation of new chemical matter. Drug discovery continues to depend on costly and largely serendipitous screening campaigns. Our understanding of, and reasoning from, non-bonded interaction chemistry is still too limited. Compounding this is the scarcity of novel chemistry and infinitesimal coverage of the chemical combinatorial space by current experimental data. To alleviate these problems, we show that a machine learning model can successfully learn and infer the principles of non-bonded interactions in the drug-receptor space. A reductionist approach to training data led to a model generalizing drug-target interactions to truly novel chemical matter without suffering from memorization bias. This work addresses that gap in drug discovery through a theoretical framework for predictive molecular recognition.