NanoBERTa-ASP: Predicting Nanobody Binding Epitopes Based on a Pretrained RoBERTa Model

Abstract

Nanobodies, also known as VHH or single-domain antibodies, are a unique class of antibodies that consist only of heavy chains and lack light chains. Nanobodies possess the advantages of both small molecule drugs and conventional antibodies, making them a promising class of therapeutic biopharmaceuticals. Studying the characteristics of nanobody sequences can aid the development and design of nanobodies. An important challenge in this field is accurately predicting the binding sites between nanobodies and antigens. The binding site is the region where the nanobody interacts with the antigen. The precise prediction of these binding sites is crucial for comprehending the interaction mechanism between the nanobody and the antigen, facilitating the design of effective nanobodies, as well as gaining valuable insights into the properties of nanobodies. Nanobodies typically have smaller and more flexible binding sites than traditional antibodies, however predictive models trained on traditional antibodies may not be suitable for nanobodies. Moreover, the limited availability of antibodyderived nanobody datasets for deep learning poses challenges in constructing highly accurate predictive models that can generalize well to unseen data. To address these challenges, we propose a novel nanobody prediction model, named NanoBERTa-ASP (Antibody Specificity Prediction), which is specifically designed for predicting nanobody-antigen binding sites. The model adopts a training strategy more suitable for nanobodies by leveraging an advanced natural language processing (NLP) model called BERT (Bidirectional Encoder Representations from Transformers). The model also utilizes a masked language modeling approach to learn the contextual information of the nanobody sequence and predict its binding site. The results obtained from training NanoBERTa-ASP on nanobodies highlight its exceptional performance in Precision and AUC, underscoring its proficiency in capturing sequence information specific to nanobodies and accurately predicting their binding sites. Furthermore, our model can provide insights into the interaction mechanisms between nanobodies and antigens, contributing to a better understanding of nanobodies, as well as accelerating the development and design of nanobodies with potential therapeutic applications. To the best of our knowledge, NanoBERTa-ASP is the first nanobody language model that achieved high accuracy in predicting the binding sites. Github repositoryhttps://github.com/WangLabforComputationalBiology/NanoBERTa-ASP