Designing Selective Drugs: Multi-Objective Optimization to Mitigate Off-Target Effects

Drug discovery is fundamental to improving human health by identifying therapeutic compounds to treat various diseases. Recent advancements in artificial intelligence (AI) have significantly accelerated this process by enabling rapid analysis of large chemical datasets and facilitating the prediction of key molecular properties for potential drug candidates. Despite these improvements, the crucial challenge of drug selectivity has received comparatively less focus in AI -driven drug discovery. In particular, drug selectivity refers to the ability of a drug to specifically bind to its intended target protein while minimizing interactions with off-target proteins, which can lead to adverse side effects. Addressing this challenge naturally involves framing it as a multi-objective optimization problem, where the goal is to simultaneously maximize target affinity and minimize off-target interactions. Traditionally, this problem has been tackled through simultaneous optimization, but this approach often struggles with high complexity due to the need to navigate the vast chemical search space in one step. To overcome this limitation, we propose a progressive optimization that first optimizes for target affinity and then focuses on minimizing off-target interactions within the constrained chemical space. We provide a theoretical explanation of our method and empirically validate its effectiveness through docking score optimization experiments to assess drug selectivity.