Deep Learning-based Design of Peptide Binders

Abstract

Peptides are short chains of amino acids that naturally mediate diverse biological functions, such as intercellular signalling, immune modulation, antimicrobial defence, and others. In recent years, peptides have also emerged as an attractive therapeutic modality. They combine best features of small molecules and biologics, enabling binding to previously undruggable molecular surfaces while remaining compact and amenable to scalable manufacturing. Yet designing peptide binders remains challenging because their interfaces are small, solvent exposed, and often highly dynamic. Recent advances in deep learning have begun to make this problem more tractable by enabling target-conditioned generation and prioritisation of peptide candidates at scale.