The Classification of Enzymes by Deep Learning

Enzymes, as a group of crucial biocatalysts produced by living cells, enable the chemical reactions in organisms to be more efficient. According to the properties of the reactions catalyzed by enzymes, the Enzyme Commission (EC) number system divided enzymes into 6 primary main classes in 1961: oxidoreductases (EC1), transferases (EC2), hydrolases (EC3), lyases (EC4), isomerases (EC5), and ligases (EC6). These six categories did not change for many years until a new class, the translocases (EC7), was added in August 2018. Different enzymes have different properties of catalytic reaction, and the prediction of enzyme classes is a very important research topic, allowing us to further study the structure and function of enzyme molecules when we know the category of enzyme. Because the number of enzymes whose function remains unknown is enormous, it is time-consuming to use biological experiments to determine enzyme characteristics. Thus, devising various computational models to predict enzyme classes has become a feasible scheme. In hope of giving researchers more inspiration and ideas for predicting the EC number of enzymes by machine learning, we summarize a variety of research methods used in the prediction of enzyme families in this research.