Property-Guided Molecular Generation and Optimization via Latent Flows

TL;DR

This paper introduces MoltenFlow, a modular framework for molecular generation and optimization that combines property-guided latent representations with flow-based models, enabling efficient multi-objective optimization and high-quality unconditional generation.

Contribution

MoltenFlow is a novel framework that integrates property-guided latent spaces with flow models for improved molecular generation and targeted optimization.

Findings

Guided latent flows enable efficient multi-objective molecular optimization.

A learned flow prior enhances unconditional generation quality.

The framework supports both conditioned generation and local optimization.

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.