Multi-view Deep Learning Multifunctional Enzyme Prediction Based on Fusion of Sequence and Multi label Embedded Information

• 1、School of Artificial Intelligence and Computer Science, Jiangnan University，Wuxi 214122

Abstract：Multifunctional enzymes can have positive effects on the survival and evolution of organisms in different functions and forms, so it is crucial to understand the functions of related enzymes. Currently, traditional machine learning methods have been widely used in enzyme function classification prediction methods, but most of them are only for the classification prediction task of single-function enzymes, and the few existing multifunctional enzyme classification models can only predict the first level of enzyme committee (EC) numbering. To address the above challenges, this paper proposes a multi-view deep learning multifunctional enzyme prediction method that fuses sequence and multi-label embedding information. In this method, a hybrid network consisting of a convolutional neural network (CNN) with attention mechanism and a bi-directional long-short memory network (Bi-LSTM) is used to learn the deep features of enzyme sequences. Meanwhile, an EC class correlation graph is constructed for the classification prediction model of each layer of EC numbering, and the EC class correlation labels are embedded using a graph convolutional network (GCN), and the obtained label embeddings are used to guide the feature learning process. Finally, a multi-label classifier is used for classification prediction of multifunctional enzymes. The experimental results show that the method achieves 75.75% accuracy for a subset of EC codes in the fourth layer, and its Macro_F1 parameter reaches 90.41%, which is a certain improvement in the prediction performance of all four EC code layers of multifunctional enzymes compared with existing methods.

Keywords： deep learning multifunctional enzyme classification multi-view Learning graph convolutional network multi-label classification