Finding Balance: Multi-Objective Optimization in Molecular Generative Modeling

Identifying novel therapeutics active towards a single target that balance requirements for potency, safety, metabolic stability and pharmacodynamic profile still presents a major challenge, which is further exacerbated by recent interest in designing compounds with properties that enable them to engage multiple targets. This entails trying balancing different, sometimes competing chemical features, which can be particularly challenging without the aid of computational methodologies. In this work, we leverage multi-objective optimization methods to help the design of novel small molecules optimised for conflicting pharmacological attributes with generative models. Across three case studies, we show that our approach is effective in generating de novo compounds predicted to have a good balance between desired properties, while also showing potential affinity to multiple targets, even when trained with limited public data. This offers a promising solution for overcoming limitations in drug discovery and design, especially when compounds with a well-balanced profile of conflicting features are needed.