Interpolation-Based Conditioning of Flow Matching Models for Bioisosteric Ligand Design

Yael Ziv, Martin Buttenschoen, Lukas Scheibelberger, Brian Marsden, Charlotte Deane

Keywords: drug discovery, 3D molecule generation, bioisosteric fragment merging, conditional generation, flow matching, generative models

Abstract:

Fast, unconditional 3D generative models can now produce high-quality molecules, but adapting them for specific design tasks often requires costly retraining. To address this, we introduce Interpolate-Integrate and Replacement Guidance, two training-free, inference-time conditioning strategies that provide control over E(3)-equivariant flow-matching models.
Our methods generate bioisosteric 3D molecules by conditioning on seed ligands or fragment sets to preserve key determinants like shape and pharmacophore patterns, without requiring the original fragment atoms to be present. We demonstrate their effectiveness on three drug-relevant tasks: natural product ligand hopping, bioisosteric fragment merging, and pharmacophore merging.

Primary Area: applications to physical sciences (physics, chemistry, biology, etc.)

Submission Number: 9488