Computational Peptide Design: Emerging Concepts in Antibiotics, Immunity, and Evolution

Abstract

Peptides are among the smallest functional biomolecules. Learning to systematically design their activity would revolutionize biology and medicine. Yet achieving this goal remains an open challenge. Recent advances in machine learning enable the discovery and design of peptides at scale, not only yielding new antibiotic agents effective against priority pathogens but also revealing unexpected roles for peptides in host immunity. In this review, we survey how artificial intelligence (AI) methods can be used to mine genomes, proteomes, and even ancient sequence data to uncover "encrypted" peptide antibiotics across the tree of life and over evolutionary time. We highlight emerging approaches for activity prediction, structure-activity modeling, sequence optimization, and de novo peptide generation, and we discuss workflows that tightly couple in silico design with high-throughput experimental validation. We further introduce the concept of "encrypted immunity," in which ordinary proteins encode latent antimicrobial fragments, and we outline translational opportunities as well as remaining challenges on the path to bringing AI-designed peptides to the clinic.