3D-EDiffMG: 3D equivariant diffusion-driven molecular generation to accelerate drug discovery

Area of Science:

• Computational chemistry and cheminformatics
• Drug discovery and development
• Artificial intelligence in medicine

Background:

• Lead compound optimization is vital for drug discovery, requiring improved biological activity and ADMET properties.
• Current deep molecular generative models struggle with scaffold modification, long-range dependencies, and molecular stability/diversity.
• Existing diffusion models lack interatomic constraint features, limiting their application in complex molecular structures.

Purpose of the Study:

• To develop an advanced deep molecular diffusion generative model for scaffold-based molecular optimization.
• To address limitations in existing models regarding interatomic constraints and long-range dependencies.
• To generate novel, stable, and diverse drug-like molecules for accelerated drug discovery.

Main Methods:

• Proposed the 3D equivariant diffusion-driven molecular generation (3D-EDiffMG) model.
• Introduced a dual strong and weak atomic interaction force-based long-range dependency capturing equivariant encoder (dual-SWLEE) for comprehensive atomic interaction encoding.
• Incorporated a gate multilayer perceptron (gMLP) block with tiny attention to model complex feature interactions and long-range dependencies.

Main Results:

• 3D-EDiffMG effectively generates unique, novel, stable, and diverse drug-like molecules.
• The model demonstrates superior performance in capturing long-range dependencies and interatomic constraints compared to existing methods.
• Experimental results validate the model's capability for scaffold-based molecular optimization.

Conclusions:

• 3D-EDiffMG offers a promising approach for accelerating lead optimization in drug discovery.
• The model's ability to generate stable and diverse molecules enhances its potential for creating new therapeutics.
• This work advances the application of deep generative models in computational drug design.