Predicting optimal catalytic temperature of enzymes using Graph Convolutional Networks


  • LUKE MEI School of Systems Biology, George Mason University, Fairfax, VA
  • Iosif Vaisman School of Systems Biology, George Mason University, Fairfax, VA



As enzymes that are stable at higher temperatures play an important role in industrial applications for enhancing reaction rates in biochemical processes, accurate prediction of optimal catalytic temperature (Topt) of enzymes can be valuable in biotechnology. Furthemore, the number of experimentally determined Topt values are limited as they are difficult and time consuming to retrieve. Existing enzyme temperature optima prediction models are based on traditional machine learning algorithms and only utilize protein sequence information. Here, we introduce DeepTOP, a Graph Convolutional Network for predicting Topt of enzymes by utilizing both protein structure as well as protein sequence features extracted using a pre-trained language model. A dataset of 2917 enzymes with experimental Topt was obtained from a previous study that produced the state of the art model. Out of the 2917 enzymes, features were extracted from 2859 with alpha fold predicted three dimensional structure. DeepTOP exhibits near state of the art performance with R^2 of 0.45 and mean squared error (MSE) of 247. 





College of Science: School of Systems Biology