A genotype-to-drug diffusion model for generation of tailored anti-cancer small molecules

Area of Science:

• Computational Biology
• Drug Discovery
• Artificial Intelligence

Background:

• Precision oncology faces challenges from tumor heterogeneity and limited drug targets.
• Generative AI offers a novel approach to design anti-cancer molecules based on genomic data.

Purpose of the Study:

• To introduce Genotype-to-Drug Diffusion (G2D-Diff), a generative AI model for designing small molecule drugs targeting specific cancer genotypes.
• To evaluate G2D-Diff's performance in generating diverse, effective, and feasible drug candidates.

Main Methods:

• Developed G2D-Diff, a generative AI model utilizing a diffusion approach.
• Trained the model on drug response data to learn genotype-specific efficacy conditions.
• Incorporated an attention mechanism for target and pathway identification.

Main Results:

• G2D-Diff generated diverse, drug-like compounds meeting specific efficacy criteria.
• The model outperformed existing methods in diversity, feasibility, and condition fitness.
• Case studies in triple-negative breast cancer identified plausible drug candidates and relevant pathways.

Conclusions:

• G2D-Diff represents a significant advancement in AI-guided personalized drug discovery.
• The model effectively combines molecule generation with pathway identification for specific genotypes.
• This approach has the potential to accelerate hit identification for challenging cancers.