Development of Neural Networks for Biomolecular Structure Prediction with Applications to Protein Design

Abstract

A grand challenge in biology is to create computational models of the interactions betweenabitrary biomolecular structures. In this dissertation, I describe the development of neural network models for predicting the structure of biomolecular complexes including proteins, nucleic acids, and small molecules. First, we developed a general neural network architecture for the prediction of biomolecular complexes in the Protein Data Bank (PDB). We then demonstrated the ability of this model to predict the structure of new complexes with high accuracy. Subsequently, we applied this model of native biomolecular complexes to the design of de novo small molecule binding proteins and enzymes. Finally, we developed a framework for development of future neural networks trained on the PDB and apply it to train several structure prediction models. To our knowledge, this dissertation represents the first efforts to develop general-purpose neural network models for biomolecular structure prediction and design.