Hybrid Diffusion Model for Stable, Affinity-Driven, Receptor-Aware Peptide Generation.

Area of Science:

• Biotechnology and Artificial Intelligence in Drug Discovery
• Computational Chemistry and Bioinformatics
• Therapeutic Peptide Design

Background:

• Peptides offer targeted therapy and fewer side effects compared to small molecules but face challenges in bioavailability and degradation.
• Advances in deep learning present opportunities for innovative peptide design.
• Targeting specific proteins, like those in *Plasmodium falciparum* erythrocyte membrane protein 1 (PfEMP1), is crucial for disease treatment.

Purpose of the Study:

• To introduce HYDRA, a novel hybrid deep learning approach for the de novo design of therapeutic peptides.
• To demonstrate HYDRA's capability in designing peptide binders for specific target receptors.
• To apply HYDRA for designing peptides targeting PfEMP1 proteins relevant to malaria.

Main Methods:

• Developed HYDRA, a hybrid deep learning model integrating diffusion models for distribution modeling.
• Combined diffusion models with a binding affinity maximization algorithm for peptide design.
• Conditioned peptide generation on target receptor binding sites for precise targeting.

Main Results:

• Successfully designed therapeutic peptides using the HYDRA approach.
• Demonstrated the model's ability to generate peptides targeting specific binding sites on receptors.
• Applied HYDRA to design peptides targeting PfEMP1, a key protein in malaria pathogenesis.

Conclusions:

• HYDRA represents a promising AI-driven strategy for therapeutic peptide design.
• The approach facilitates the development of targeted peptide therapeutics for diseases such as malaria.
• This method holds potential for advancing drug discovery beyond malaria to other diseases.