Guided diffusion for molecular generation with interaction prompt.

Area of Science:

• Computational chemistry
• Drug discovery
• Molecular modeling

Background:

• Molecular generative models aid in designing molecules for specific protein pockets.
• Current models lack control over the generative process and ignore crucial ligand-protein interactions.
• Hotspot residues are key targets due to their significant contribution to binding energy.

Purpose of the Study:

• To develop a guided molecular generative model that incorporates atomic interaction information.
• To address the randomness and lack of specificity in existing generative models.
• To enable the design of molecules with desired interactions with protein hotspots.

Main Methods:

• Development of InterDiff, an interaction prompt-guided diffusion model.
• Representation of four atomic interaction types as learnable vector embeddings.
• Utilizing these embeddings as conditions to guide residue-specific molecular generation.

Main Results:

• InterDiff successfully generates molecules with desired ligand-protein interactions in a controllable manner.
• In silico experiments demonstrate the model's effectiveness.
• Validation on two therapeutic agents shows InterDiff generates molecules with comparable or superior binding modes to existing drugs.

Conclusions:

• InterDiff offers a novel approach to structure-based drug design by guiding molecular generation with specific interactions.
• The model shows potential for designing targeted therapeutics with improved binding affinities and modes.
• This work advances the field of AI-driven drug discovery by enabling more precise molecular design.