Harnessing Deep Learning for the Rational Design of Metalloproteins

This thesis project aims to explore the application of Deep Learning in modeling and generating de novo metalloproteins. The investigation will be organized into three key components. Firstly, a Graph Neural Network will undergo training using metal-based small molecules, leveraging the extensive Cambridge Structural Database (CSD) repository with validated crystal structures containing metal sites. This model is designed to comprehend and capture the intricate relationship between 3D structure and metal coordination. Secondly, the knowledge acquired from small molecules will be translated to adapt the framework for the protein domain. Lastly, a structural generative model will be crafted to predict the 3D structure of metalloproteins, with the assistance of large language models to refine the encoding sequences of the generated metalloproteins.