Prediction and Design of Protease Specificity Using a Structure-Aware Graph Convolutional Network

Abstract

Translate Article

Specific molecular recognition of substrates by enzymes, <em>e.g.</em>, proteases, is critical for maintaining the robustness of crucial life processes. The ability to predictably tailor protease specificity would enable targeted proteolytic cleavage of any chosen target protein. Current methods for predicting protease specificity rely on sequence pattern recognition in experimentally-derived cleavage data obtained for libraries of substrates. We reasoned that a more semantically rich and robust model of protease specificity could be developed by incorporating the three-dimensional structure and energetics of molecular interactions between protease and substrates. We present a Protein Graph Convolutional Network (PGCN), which uses a physically-grounded, structure-based molecular interaction graph representation that describes molecular topology and energetics to predict specificity. PGCN was shown to accurately predict the specificity landscapes of several variants of two model proteases: the NS3/4 protease from the Hepatitis C virus (HCV) and the Tobacco Etch Virus (TEV) proteases. Node and edge ablation tests identified key graph elements for specificity prediction, some of which are consistent with known biochemical constraints for protease:substrate recognition. We used a pre-trained PGCN model to guide the design of TEV protease libraries for cleaving two non-canonical substrates and found high correspondence between PGCN prediction and experimental cleavage results for 19 designed protease variants. The described methodology should be applicable to the construction of tailor-made proteases for site-selectively cleaving chosen substrate sequences, enabling broad applications in protein design. Here is a brief description of each file: protease-gcnn-pytorch.zip PGCN codes. See the latest version here: https://github.com/Nucleus2014/protease-gcnn-pytorch train_val_test_graph_datasets.zip Datasets for the training process of PGCN. Statistics of them are shown in Table 1. tev_design_for_validation_pdbstructures.zip Final selected designs for experimental testing tev_oydv_design_candidates_pdbstructures.zip Candidate designs made by Rosetta