When Trees Guide Molecules: Multiobjective Search in de Novo Drug Design

Seen as a problem of guided exploration with multiple objectives, molecular design requires carefully elaborated decision-tree-based strategies for traversing vast chemical space. Among main challenges, the discovery of bioactive molecules with either selective or polypharmacological inhibitory profiles remains a central one. In this work, we present a computational framework that integrates Monte Carlo tree search with predictive machine learning models and expert-defined rules to navigate de novo generation of inhibitor candidates toward orexin and histamine receptors. To showcase such a multiobjective optimization, besides the inhibitory action, the candidate molecules are also optimized against a number of properties─toxicity, aqueous solubility, and melting temperature. This hybrid approach enables efficient exploration of chemical space toward molecules with desired selectivity or multitarget activity profiles, offering a scalable path for in silico drug design.