Diff-Shape: A Novel Constrained Diffusion Model for Shape based De Novo Drug Design

Mingyuan Xu,Hongming Chen,Jie Lin

DOI: https://doi.org/10.26434/chemrxiv-2024-km0h1

2024-04-30

Abstract:Shape-based virtual screening is a widely utilized method in ligand-based de novo drug design, aiming to identify molecules in chemical libraries that share similar 3D shapes but simultaneously possess novel 2D chemical structures compared to the reference compound. As an emerging technology, generative model is an alternative way to do de novo drug design by directly generating 3D novel structures. However, existing models face challenges in reliably generating valid drug-like molecules under specific conformation constrains. Here, a novel diffusion model constrained with 3D reference shape, Diff-Shape, was proposed to generate structures whose 3D conformations are similar to a given reference shape, thereby avoiding the computational cost of screening large database of 3D conformations. This model utilized a zero-weighted graph control module, taking in various forms of point clouds of reference shape to guide diffusion process of 3D molecular generation. The results show that our model is capable of generating molecules with high shape similarity but still low 2D graph similarity to the query structure and it significantly out-performs existing shape based generative models. Examples were also given to demonstrate that combining with docking software our model can effectively generate structures in structure-based drug design scenario.

Chemistry

What problem does this paper attempt to address?

The paper aims to address a problem in shape-based de novo drug design, specifically how to generate new molecules that have similar 3D conformations to a given reference shape but possess novel 2D structures. Existing generative models face challenges in reliably generating drug-like molecules under specific conformation constraints. To solve this problem, the paper proposes a new diffusion model, Diff-Shape, which is constrained by a 3D reference shape to generate molecules similar to the reference shape, thereby avoiding the computational cost of screening large 3D conformation databases. Additionally, the model demonstrates effectiveness in structure-based drug design scenarios when combined with docking software. Experimental results show that the Diff-Shape model significantly outperforms existing shape-based generative models in generating molecules with high shape similarity and low 2D graph similarity.