ProtNN: Fast and Accurate Nearest Neighbor Protein Function Prediction based on Graph Embedding in Structural and Topological Space

摘要

Studying the function of proteins is important for understanding themolecular mechanisms of life. The number of publicly available proteinstructures has increasingly become extremely large. Still, the determination ofthe function of a protein structure remains a difficult, costly, and timeconsuming task. The difficulties are often due to the essential role of spatialand topological structures in the determination of protein functions in livingcells. In this paper, we propose ProtNN, a novel approach for protein functionprediction. Given an unannotated protein structure and a set of annotatedproteins, ProtNN finds the nearest neighbor annotated structures based onprotein-graph pairwise similarities. Given a query protein, ProtNN finds thenearest neighbor reference proteins based on a graph representation model and apairwise similarity between vector embedding of both query and referenceprotein-graphs in structural and topological spaces. ProtNN assigns to thequery protein the function with the highest number of votes across the set of knearest neighbor reference proteins, where k is a user-defined parameter.Experimental evaluation demonstrates that ProtNN is able to accurately classifyseveral datasets in an extremely fast runtime compared to state-of-the-artapproaches. We further show that ProtNN is able to scale up to a whole PDBdataset in a single-process mode with no parallelization, with a gain ofthousands order of magnitude of runtime compared to state-of-the-artapproaches.