DiffBP: Generative Diffusion of 3D Molecules for Target Protein Binding

Abstract

Generating molecules that bind to specific proteins is an important but challenging task in drug discovery. Most previous works typically generate atoms autoregressively, with element types and 3D coordinates of atoms generated sequentially. However, in real-world molecular systems, atomic interactions span the entire molecule, necessitating a consideration of pairwise-additive energy among atoms. Given this energy-centric perspective, modeling probability should rely on joint distributions instead of sequential conditional ones. Thus, the conventional sequential auto-regressive methods for molecule generation can inadvertently violate physical principles, yielding molecules with undesirable properties. In this study, we propose DiffBP, a generative diffusion model that generates 3-dimensional (3D) molecular structures, leveraging target proteins as contextual constraints at the full-atom level in a non-autoregressive way. When provided with a specified 3D protein binding site, our model learns to denoise both the element types and 3D coordinates of the entire molecule using an equivariant network. Experimental assessments illustrate that DiffBP performs competitively against existing methods, generating molecules with high protein affinity, appropriate molecule sizes, and desirable drug-like profiles. Additionally, we develop a website server for medicinal chemists interested in exploring the art of molecular generation, which is accessible at http://www.manimer.com/moleculeformation/index.