Computer-aided epitope prediction of target specific protein binders

Abstract

Protein binders, represented by the antibody, are one of the most effective therapeutics for a wide range of target diseases. Therapeutic protein binders, in general, specifically interact with the biomarker of the pathogen and neutralize them. In terms of understanding the molecular mechanism of neutralization, it is important to localize the binding epitope. Whereas the advances in in vitro library display screening and next-generation sequencing have enabled accelerated development of strong binders, yet identifying their binding sites still remains a major challenge. In this regard, we developed two different computational methods to aid in epitope characterization of protein binders: computational epitope binning and epitope prediction. In chapter 1, we introduce a computational epitope binning method ‘Epibin’ which takes sequences of protein binders as input to generate homology models of each binder and utilize them for docking against the target protein. Epibin considers the generated docking models (even decoys along with near-native models) as a fingerprint of each binder, which would represent the physicochemical characteristics of the paratope. We scored each pair of protein binders for the similarity (or competition) of binding based on the generated docking models and classified them into separate bins of epitopes. Epitope binning is particularly valuable in protein binder development, when a large number of protein binders are screened for a single target and the pool has to be narrowed down to a handful of leads with similar traits (epitopes). As a result, we have successfully shown that Epibin can segregate binders with different epitopes and group together binders with a nearby epitope, using one set of repebodies targeting interleukin-6, and two sets of antibodies each targeting SARS-CoV-2 and Pfs25. Chapter 2 shows a computational epitope prediction technique using molecular dynamics simulations to extract a local flexibility of a target protein. Here, we compared the structural flexibility of epitopes and non-epitopes of target proteins from a non-redundant set of protein binders including antibody, DARPin, monobody, affibody and repebody. The results indicate that, in general, the epitope of a target protein is less structurally flexible compared to the non-epitope surface. We also show that this tendency is stronger when the paratope of the protein binder is mainly consisted of β-sheets (e.g. monobody, repebody). We utilized this information to design an epitope prediction tool for a target protein based on its local flexibility, which showed a comparable performance with some of the most widely used B-cell epitope prediction tools. In this study, we demonstrated the utility of two computational tools for epitope binning or epitope prediction. These approaches can be utilized in protein binder development, especially in therapeutics development, where a large pool of protein binders need to be evaluated in a high-throughput manner. Furthermore, the trends of protein binding preference and secondary structure of a protein may also be helpful in designing protein-protein interactions.