Guided multi-objective generative AI to enhance structure-based drug design.

Area of Science:

• Computational chemistry and artificial intelligence in drug discovery.

Background:

• Current generative AI models struggle to create molecules with desired physicochemical properties for drug design.
• Structure-based drug design requires optimizing multiple parameters like binding affinity and synthetic accessibility.

Purpose of the Study:

• To introduce IDOLpro, a novel generative chemistry AI platform for structure-based drug design.
• To demonstrate IDOLpro's ability to generate molecules with optimized binding affinity and synthetic accessibility.

Main Methods:

• IDOLpro combines diffusion models with multi-objective optimization.
• Differentiable scoring functions guide the diffusion model's latent variables to explore chemical space.
• The platform generates novel ligands in silico, optimizing multiple physicochemical properties.

Main Results:

• IDOLpro generated ligands with 10-20% higher binding affinity than state-of-the-art methods on benchmark sets.
• The platform produced more drug-like molecules with better synthetic accessibility scores.
• IDOLpro outperformed exhaustive virtual screening, generating molecules with superior binding affinity and synthetic accessibility faster and cheaper.
• IDOLpro generated molecules with higher binding affinities than experimentally observed ligands on a test set of experimental complexes.

Conclusions:

• IDOLpro represents a significant advancement in generative chemistry for drug discovery.
• The platform can accelerate hit-finding, hit-to-lead, and lead optimization by accommodating various scoring functions (e.g., ADME-Tox).
• IDOLpro is capable of generating novel drug candidates with improved properties, potentially revolutionizing the drug discovery pipeline.