Deep mutational scanning and machine learning uncover antimicrobial peptide features driving membrane selectivity.

Antimicrobial peptides are critical actors of the innate immune system. Those with direct antibacterial activity often kill via cell membrane lysis, but have difficulty distinguishing between target cells and mammalian cells complicating therapeutic use. Here, we use a new technique called deep mutational surface localized antimicrobial display (dmSLAY) to rapidly elucidate single residue importance and flexibility for Protegrin-1, a potent yet toxic host-defense peptide. Subsequent biochemical analyses reveal that bacterial membrane selectivity is improved for Protegrin-1 variants which maintain membrane bound secondary structure, avoid large aromatic residues, and mutate cysteine pairs. Machine learning is used to expand our analysis to encompass over five million sequence variations and uncover full mutational profiles which promote antibacterial potency, toxicity, and specificity for bacterial or mammalian cell membranes. Our results describe an innovative, high-throughput approach for rapidly elucidating antimicrobial peptide sequence-structure-function relationships which can be used to inform future clinical antibiotic design.