Diffusion Models in De Novo Drug Design.

Area of Science:

• Computational chemistry and cheminformatics
• Artificial intelligence in drug discovery

Background:

• Diffusion models, inspired by statistical physics, excel at learning complex probability distributions for 3D molecular geometries.
• These models are crucial for generating 3D molecular structures with desired chemical and physical properties, essential for drug discovery.

Purpose of the Study:

• To review the technical implementation of diffusion models for 3D molecular generation.
• To compare performance, evaluation methods, and implementation details of various diffusion models.
• To explore applications in de novo drug design and computational chemistry.

Main Methods:

• Focus on technical aspects: atom/bond representation, denoising network architectures.
• Comparison of different diffusion models for molecular generation tasks.
• Review of conditional generation strategies and fragment-based design.

Main Results:

• Diffusion models demonstrate success in learning and generating 3D molecular structures.
• Key strategies and challenges in implementing these models are discussed.
• Applications span de novo design, structure-based design, docking, and molecular dynamics.

Conclusions:

• Diffusion models are pivotal in advancing 3D molecular generation for drug discovery.
• The review highlights their current capabilities and limitations in computational chemistry.
• Future directions involve enhancing stability and expanding applications in rational drug design.