Hot-Spot-Guided Generative Deep Learning for Drug-Like PPI Inhibitor Design

Area of Science:

• Computational chemistry
• Drug discovery
• Artificial intelligence in medicine

Background:

• Protein-protein interactions (PPIs) are critical therapeutic targets.
• Developing small-molecule inhibitors for PPIs is challenging due to their large, flat interfaces.
• Existing computational methods are limited by chemical libraries and heuristics, restricting novel drug design.

Purpose of the Study:

• To introduce Hot2Mol, a generative deep learning framework for de novo design of PPI inhibitors.
• To enable precise targeting of PPI interfaces using pharmacophoric features from hot-spot residues.
• To overcome limitations of traditional methods and explore novel chemical space for drug discovery.

Main Methods:

• Hot2Mol integrates a conditional transformer for property-constrained molecular generation.
• An E(n)-equivariant graph neural network ensures spatial alignment with PPI hot-spot pharmacophores.
• A variational autoencoder is used for sampling diverse and novel molecular structures.

Main Results:

• Hot2Mol outperforms state-of-the-art models in binding affinity, drug-likeness, and novelty.
• Generated compounds exhibit strong binding stability, confirmed by molecular dynamics simulations.
• Case studies demonstrate the design of high-affinity and selective PPI inhibitors.

Conclusions:

• Hot2Mol effectively designs target-specific and drug-like PPI inhibitors.
• The framework accelerates rational drug discovery for challenging PPI targets.
• Hot2Mol represents a significant advancement in computational drug design for PPIs.