Property-Guided Molecular Generation and Optimization via Latent Flows

2026-03-27

AI Summary

This work addresses the challenge in molecular inverse design of simultaneously achieving desired properties, molecular validity, structural fidelity, and optimization stability. To this end, the authors propose MoltenFlow, a unified framework that integrates property-aligned representations, flow-matching generative priors, and gradient-guided optimization within a shared latent space. MoltenFlow is the first method to unify high-quality unconditional generation with controllable multi-objective conditional optimization. Experimental results demonstrate that, under a fixed evaluation budget, MoltenFlow significantly improves the validity, diversity, and optimization efficiency of generated molecules while maintaining robust stability and practical applicability.

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.

Problem

• molecular generation
• inverse design
• latent optimization
• property-guided
• chemical validity

Innovation

• latent flows
• molecular generation
• property-guided optimization
• flow matching
• multi-objective optimization