Balancing exploration and exploitation in de novo drug design

Goal-directed molecular generation is the computational design of novel molecular structures optimised with respect to a given scoring function. While it holds great promise for the acceleration of drug design, there remain limitations that hamper its adoption in an industrial context. In particular, the lack of diversity of molecules generated currently limits their relevance for drug design. Yet, most algorithms proposed focus solely on optimizing the scoring function, and do not address the question of diversity of the solutions they propose. Here, we propose a conceptual framework for analyzing the need for diverse solutions in goal-directed generation. Using a mean-variance framework, we present a simple model to bridge the optimization objective of goal-directed generation with the need for diverse solutions. We also show how to integrate it within different goal-directed learning algorithms.