[2603.26889] Property-Guided Molecular Generation and Optimization via Latent Flows

Computer Science > Machine Learning arXiv:2603.26889 (cs) [Submitted on 27 Mar 2026] Title:Property-Guided Molecular Generation and Optimization via Latent Flows Authors:Alexander Arjun Lobo, Urvi Awasthi, Leonid Zhukov View a PDF of the paper titled Property-Guided Molecular Generation and Optimization via Latent Flows, by Alexander Arjun Lobo and 2 other authors View PDF HTML (experimental) Abstract:Molecular discovery is increasingly framed as an inverse design problem: identifying molecular structures that satisfy desired property profiles under feasibility constraints. While recent generative models provide continuous latent representations of chemical space, targeted optimization within these representations often leads to degraded validity, loss of structural fidelity, or unstable behavior. We introduce MoltenFlow, a modular framework that combines property-organized latent representations with flow-matching generative priors and gradient-based guidance. This formulation supports both conditioned generation and local optimization within a single latent-space framework. We show that guided latent flows enable efficient multi-objective molecular optimization under fixed oracle budgets with controllable trade-offs, while a learned flow prior improves unconditional generation quality. Comments: Subjects: Machine Learning (cs.LG); Materials Science (cond-mat.mtrl-sci) Cite as: arXiv:2603.26889 [cs.LG]   (or arXiv:2603.26889v1 [cs.LG] for this version)   https://doi.org/1...