Antibody structure prediction using deep learning

Thesis

The aim of this doctoral project was to improve the prediction of antibody structures and optimise the use of existing antibody data. As a starting point, we developed ABlooper, a tool that predicts the position of backbone atoms for CDR loops in antibodies. In Chapter 2, we present the methods and results for ABlooper. This tool has enabled the modelling of antibody structures at previously unprecedented speeds, predicting the CDR loops for one hundred structures in less than five seconds. However, ABlooper has some limitations: it cannot model side-chain atoms and sometimes generates unphysical structures, which need to be corrected using computationally expensive methods.

To overcome these limitations, we developed ImmuneBuilder, a novel method based on the recently released AlphaFold2 models. ImmuneBuilder is a set of deep learning-based tools for modelling antibodies, nanobodies, and TCRs. In contrast to ABlooper, this method can accurately predict the position of both backbone and side-chain atoms. Moreover, it predicts the structure of the entire variable domain, not just that of the CDRs. Chapter 3 describes the ImmuneBuilder method in detail.

With next-generation sequencing studies routinely generating millions of antibody sequences, effectively searching this data has become a challenge. In Chapter 4, we present KA-Search a tool that allows for the rapid search of billions of antibody sequences by sequence identity across either the whole chain, the complementarity-determining regions, or a user defined antibody region. We show KA-Search in operation on the 2.4 billion antibody sequences available in the OAS database.

In addition to next-generation sequencing data, there is a vast amount of literature that focuses on a small number of antibodies. For these smaller scale experiments, the amount of available information for each individual antibody is much greater. To facilitate the retrieval of this information, we developed the Patent and Literature Antibody Database (PLAbDab), an evolving reference set of functionally diverse, literature-annotated antibody sequences and structures. Chapter 5 describes how this database was generated and gives examples of how it could be used.

The final chapter draws conclusions and proposes future avenues of research.

Authors: Abanades, B

• DOI: 10.5287/ora-avwodkrvv
• Type of award: DPhil
• Level of award: Doctoral
• Awarding institution: University of Oxford
• Language: English
• Keywords: antibody; ka-search; deep learning; plabdab; immunebuilder; structure; ablooper; bioinformatics
• Subjects: Immunoinformatics; Protein structure prediction; Machine Learning