TarMGDif: Target-Specific Molecular Graphs Generation Based on Diffusion Model.

Area of Science:

• Computational chemistry
• Drug discovery
• Artificial intelligence in medicine

Background:

• Generating specific drug molecules for target proteins is complex and resource-intensive.
• Existing diffusion models often require 3D information and equivariance, posing computational challenges.
• Accurate feature extraction from target proteins is crucial but difficult for many methods.

Purpose of the Study:

• To propose TarMGDif, a novel target-specific molecular graph generation model.
• To address limitations of existing 3D-based diffusion models by utilizing a discrete denoising diffusion framework for graph structures.
• To enhance the generation of chemically valid molecules with specific target-binding properties.

Main Methods:

• Developed TarMGDif, a discrete denoising diffusion model operating on molecular graphs.
• Incorporated a global features embedding network to capture ring features and learn diffusion time steps.
• Introduced a novel node-to-edge attention module to model interdependencies.

Main Results:

• TarMGDif demonstrated superior performance across three experimental datasets.
• The model successfully generated novel molecules targeting the DRD2 protein via transfer learning.
• Generated molecules exhibited pharmacological properties comparable to known DRD2 inhibitors.

Conclusions:

• TarMGDif offers an efficient and effective approach for target-specific molecular graph generation.
• The model's ability to handle graph structures and chemical validity is a significant advancement.
• TarMGDif shows strong potential for accelerating the design of novel therapeutics.