Developing Engineering Principles for Computational Peptide Design

Abstract

Peptides represent an attractive class of molecular recognition elements for many applications including therapeutics, diagnostics, and sensing because they combine favorable properties of both small molecules and larger proteins. However, their inherent conformational flexibility complicates the prediction of their structures and, subsequently, their interactions, limiting the development of generalizable computational design strategies. Although recent advances in machine learning have transformed the computational protein field and substantially improved the prediction and design of proteins with stable structures, these advances have not translated as effectively to peptides. Determining which structural representation best describes the bound conformation of a peptide remains a significant challenge, and this uncertainty propagates throughout computational design workflows. This structural ambiguity and the fragmented availability of experimental thermodynamic data for protein-peptide complexes limit mechanistic insight and continue to constrain computational approaches for engineering peptides. Therefore, advancing rational peptide design requires experimentally grounded frameworks capable of revealing the molecular principles underlying peptide recognition. In this work, computational approaches were used to investigate the determinants of protein-peptide binding and develop engineering principles for peptide design. Collaborative evaluation of a previously reported insulin-binding peptide demonstrated that candidate recognition elements require rigorous validation and highlighted the need for improved computational frameworks to understand peptide binding. To address the fragmented availability of peptide-binding data, the Predicted and Experimental Peptide Binding Information (PEPBI) database was developed by pairing experimentally determined thermodynamic measurements with curated structural models and computational descriptors of protein-peptide complexes. This resource enabled systematic investigation of the relationships between molecular features and binding thermodynamics across diverse peptide systems. Analyses of the PEPBI database enabled the development of a framework for classifying peptide residues according to their energetic importance and revealed recurring interface- and residue-level features associated with favorable binding behavior. Those same molecular features were subsequently used to develop predictive models relating them to changes in binding free energy (ΔΔG). Although these models exhibited limited generalization across previously unseen protein-peptide systems, they demonstrated utility as practical tools for prioritizing candidate peptides for experimental evaluation. Together, these findings establish engineering principles relevant to rational peptide design and support the development of peptide-based recognition elements within the broader objectives of the BIO-SENS initiative.