Computational Prediction of Protein-Peptide Binding

Abstract

Protein-peptide interactions are crucial for many important biological processes, especially during signal transduction, as they regularly trigger signaling by initiating molecular recognition events. Additionally, peptides are natural inhibitors for proteins and therefore important lead structures in pharmaceutical research. Prominent examples for peptide-based drugs are inhibitors of viral proteases [1, 2].There exist very few computational approaches, which allow a structure-based prediction of protein-peptide binding, especially for larger peptides (> 5 amino acids) and surface-exposed binding sites. We have developed a two-stage method for this purpose.During our procedure, we first predict the peptide's binding site on the protein's surface. This step is important, as for many biologically relevant protein-peptide interactions no structural information is available for the bound complex. Afterwards we sample all possible peptide conformations in the predicted binding site to identify the bound conformation of the protein-peptide complex using two methods: IRECS [3, 4], an algorithm for predicting side chain conformations, and DynaDock[5], a molecular dynamics-based docking approach, which allows an efficient description of the protein's flexibility during protein-peptide assembly.The methodology was successfully evaluated in various projects e.g. investigating peptide binding to Hsp70 and MHC proteins [6]. We will give an outline of the methodology and its application.