Graphs are correct representations of proteins to understand their structures and extract their functional characteristics. However, classical machine learning models cannot be used directly on graphs for such purposes. To benefit from graphs in machine learning modeling of proteins, we used graph neural networks. In a graph of a protein, residues are considered as the nodes and edges are defined based on neighbourhood definition using residue distance in the Cartesian coordinates system. Our graph neural network model is designed to predict binary classification tasks on protein residue level. Using this approach we could achieve residue level information retrieval for human proteins with high performance.

Understanding residue level protein chemistry using graph neural networks

ACS Fall 2021

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.