Classification And Recognition Of Enzymes By Graph Neural Network

Enzymes play a very important role in people's production and life.In August 2018,International Federation of Biochemistry and Molecular Biology changed the classification rules of enzymes,and added a new enzyme,Translocases,to the original six enzymes,numbered EC7.With the discovery of more and more enzymes with new functions and the generation of many new protein in the protein project,these protein will be included in the existing protein database,and the number of protein in the protein database is constantly increasing.Choosing effective classification and recognition methods in a huge protein database plays an important role in enzyme classification.In this paper,a new feature is considered in the training of graph neural network model,which is the spatial structure feature between amino acids in enzyme structure.In the model training tasks of machine learning and deep learning,feature acquisition is a very important task.This paper mostly considers the new feature of amino acid spatial structure in enzyme structure,which is used for enzyme classification and recognition tasks in graph neural network training.In previous studies on enzyme classification,researchers generally only considered the node feature information of amino acids in the process of model training,but did not consider the topological structure feature information of amino acids.Graph neural network has become a research hotspot in the field of artificial intelligence in recent years,which provides a new direction for classification and recognition.The characteristics of amino acid nodes and amino acid topological structure in enzyme protein structure are proposed comprehensively in this paper,and the model is trained by graph neural network.The main research contents are as follows: Firstly,this paper extracts the node information features of amino acids and the topological structure information features between amino acids from the data set,and selects some data as the input of graph neural network to train the model,and uses the remaining data as test data to verify the accuracy of classification by graph neural network method;Secondly,this paper uses K nearest neighbor algorithm,support vector machine,random forest algorithm and multi-layer perceptron to make comparative experiments with the same data.Finally,the receiver operator characteristic curve,accuracy,sensitivity and specificity of the subjects are obtained by these five methods.After comparative analysis,it is concluded that the comprehensive effect of neural network method is more excellent.