CProMG: controllable protein-oriented molecule generation with desired binding affinity and drug-like properties

Area of Science:

• Computational chemistry
• Drug discovery
• Artificial intelligence in medicine

Background:

• Deep learning accelerates exploration of chemical space for de novo molecule design.
• Generating molecules with specific protein binding affinities and drug-like properties remains a challenge.

Purpose of the Study:

• To introduce CProMG, a novel framework for controllable protein-oriented molecule generation.
• To enhance protein binding pocket representation by fusing hierarchical protein views.
• To enable autoregressive generation of molecules with controlled binding affinity and drug-like properties.

Main Methods:

• Developed CProMG with a 3D protein embedding module, dual-view protein encoder, molecule embedding module, and drug-like molecule decoder.
• Fused hierarchical protein views, associating amino acid residues with atoms for enhanced binding pocket representation.
• Jointly embedded molecule sequences, drug-like properties, and protein binding affinities for controllable generation.

Main Results:

• CProMG demonstrated superior performance compared to state-of-the-art deep generative methods.
• Progressive control of binding affinity and drug-like properties was achieved, validating the framework's effectiveness.
• Ablation studies confirmed the contributions of hierarchical protein views, Laplacian position encoding, and property control.

Conclusions:

• CProMG offers a novel and effective approach for controllable protein-oriented molecule generation.
• The framework enhances de novo molecule design by capturing crucial protein-molecule interactions.
• This work is anticipated to significantly boost advancements in drug discovery.