Finding Balance: Multiobjective Optimization in Molecular Generative Modeling

Identifying novel therapeutics active against a single target that balance the requirements for potency, safety, metabolic stability, and a favorable pharmacodynamic profile remains a major challenge, further exacerbated by recent interest in designing compounds with properties that enable engagement of multiple targets. Computational methods such as predictive modeling and generative design can efficiently navigate chemical space to find such candidates. In this work, we leverage multiobjective optimization methods, together with generative models, to aid the design of novel small molecules optimized for conflicting pharmacological attributes. Across three case studies, we show that our approach is effective at generating de novo compounds predicted to have a favorable balance of desired properties and to exhibit potential affinity for multiple targets, even when trained on limited public data. Ultimately, this approach offers a practical strategy for identifying de novo compounds that satisfy complex, multiobjective profiles in applied drug design.