MolSculptor: An Adaptive Diffusion-Evolution Framework Enabling Generative Drug Design for Multitarget Affinity and...

Area of Science:

• Computational chemistry and drug discovery.
• Artificial intelligence in medicine.
• Molecular modeling and design.

Background:

• Designing molecules for multiple protein targets is vital for complex diseases like cancer but faces significant challenges.
• Deep generative models struggle with multi-target structural information and require extensive data or expert knowledge.
• Existing methods lack the flexibility to design inhibitors for arbitrary target combinations.

Purpose of the Study:

• To introduce MolSculptor, an adaptive diffusion-evolution framework for generating inhibitors against any combination of on- and off-targets.
• To circumvent the need for target-specific training data or prior expert knowledge in drug design.
• To provide a versatile workflow for both de novo design and lead optimization, adaptable to various drug discovery stages.

Main Methods:

• Utilizes a 3D-aware surrogate model for flexible guidance based on specified targets.
• Employs an active learning protocol for adaptive refinement of guidance, ensuring performance in data-scarce scenarios.
• Integrates de novo design and lead optimization with direct conditioning on drug-like properties.

Main Results:

• MolSculptor significantly outperforms state-of-the-art methods in multitarget and selective inhibitor design tasks.
• Generated molecules demonstrate predicted affinity profiles superior to experimentally validated references.
• Successfully designed and synthesized a novel dual-target inhibitor for castration-resistant prostate cancer (CRPC), confirmed by wet-lab validation.

Conclusions:

• MolSculptor offers a powerful and generalizable paradigm for designing ligands with complex, multitarget activity profiles.
• The framework enables data-efficient solutions for complex therapeutic problems.
• Paves the way for accelerated development of therapies for challenging diseases.