DrugHIVE: Target-specific spatial drug design and optimization with a hierarchical generative model.

Area of Science:

• Computational chemistry and cheminformatics
• Artificial intelligence in drug discovery
• Structural biology and molecular modeling

Background:

• Structure-based drug design (SBDD) is crucial in early drug development, leveraging advancements in computational methods and structural data.
• Virtual screening methods are impactful but face scalability challenges due to vast chemical space and computational limits.
• Deep generative modeling offers a powerful AI-driven approach to overcome limitations by learning molecular relationships from data.

Approach:

• Introduced DrugHIVE, a deep hierarchical structure-based generative model for controlled molecular generation.
• The model learns intra- and inter-molecular relationships within drug-target systems from existing data.
• DrugHIVE is designed for scalability and fine-grained control over the generated molecular structures.

Key Points:

• DrugHIVE outperforms state-of-the-art autoregressive and diffusion-based generative models in benchmarks and generation speed.
• Demonstrated efficacy across diverse drug design tasks: de novo generation, molecular optimization, scaffold hopping, linker design, and pattern replacement.
• Successfully applied to high-confidence AlphaFold predicted receptors, expanding drug-like molecule generation to previously inaccessible targets.

Conclusions:

• DrugHIVE significantly accelerates common drug design workflows, offering enhanced control and efficiency.
• The model's scalability and applicability to predicted protein structures broaden the scope of AI-driven drug discovery.
• Enables high-quality drug-like molecule generation for a majority of the human proteome, including targets with unsolved structures.