PepINVENT: generative peptide design beyond natural amino acids

Area of Science:

• Medicinal Chemistry
• Artificial Intelligence in Drug Discovery
• Computational Chemistry

Background:

• Peptides are vital in drug design, acting as therapeutics or delivery agents.
• Non-natural amino acids (NNAAs) improve peptide properties like stability and permeability.
• Current generative models are limited to predefined amino acid sets, hindering exploration of novel chemical space.

Purpose of the Study:

• To introduce PepINVENT, a generative AI tool for exploring the theoretical chemical space of peptides.
• To enable the de novo design of novel amino acids and peptides beyond existing libraries.
• To facilitate goal-oriented peptide design and property optimization.

Main Methods:

• PepINVENT, an extension of the REINVENT platform, utilizes generative AI and reinforcement learning.
• The model was trained to understand peptide granularity and design amino acids for masked positions.
• It navigates natural and non-natural amino acid space to propose valid, novel, and diverse peptide designs.

Main Results:

• PepINVENT successfully explores peptide space, generating unique and novel designs.
• The tool demonstrates capacity for property optimization in therapeutically relevant peptides.
• The generative model is versatile, not limited by specific peptide properties or topologies.

Conclusions:

• PepINVENT offers a powerful approach to explore uncharted peptide chemical space.
• The tool supports multi-parameter learning, peptidomimetics, and lead optimization in peptide drug discovery.
• PepINVENT facilitates the design of more effective peptides with enhanced properties.