De novo generation of peptide binders with desired properties by deep generative models reinforced through enrichment of focused sets for iterative fine-tuning

Abstract

Recurrent neural networks underwent reinforcement procedures for de novo generation of peptide binders with desired properties. Docking and scoring of peptides from these models allowed enrichment of focused sets with validated sequences for iterative fine-tuning, leading to reinforcement of those models. They enabled de novo generation of peptide sequences with high binding affinity to the target and possibly additional properties.