Research On A New Intelligent Prediction Method For Multifunctional Enzymes

A multifunctional enzyme is a special type of enzyme that catalyzes a variety of basic chemical reactions.Studies have shown that multifunctional enzymes can catalyze different chemical reactions in different forms,which makes the research value and application value of multifunctional enzymes higher than that of ordinary single-function enzymes.Traditional methods of enzyme function research mostly use biological experimental techniques such as enzyme analysis.For the classification of multifunctional enzymes and the function determination of the rapidly increasing number of new enzymes,biotechnology-based methods are time-consuming and expensive.In order to cope with the above challenges,computational methods such as machine learning have been attempted to deal with the problem of enzyme function classification in recent years.The multifunctional enzyme prediction problem is essentially a multi-label learning problem.At present,some classification and prediction methods of multifunctional enzymes based on machine learning have been proposed.Although the existing classification and prediction methods of multifunctional enzymes have shown certain effectiveness,there are still two important challenges as follows.On the one hand,almost all of the existing methods only consider the sequence features of the enzyme,and do not make full use of the label features of the class.On the other hand,most of the existing methods for predicting multifunctional enzymes only consider the prediction of the major classes of multifunctional enzymes,but fail to achieve the prediction of the complete EC coding of the multifunctional enzymes.In response to the above challenges,this paper conducts an in-depth study and proposes two novel multifunctional enzyme classification prediction models.Specifically,the main work of this paper can be summarized as follows:(1)In order to make full use of the label semantic features of enzyme categories and fully predict the complete EC code of multifunctional enzymes,a multi-view deep learning multifunctional enzyme prediction method ml DGCnet(Multi-label Deep learning GCN-CNN net)combining sequence and multi-label embedding information was proposed.The method introduces multi-view learning,multi-label learning mechanism and graph convolutional deep learning network structure,and constructs multi-view feature set by extracting sequence correlation features and sequence independent features of enzyme sequences.At the same time,the graph convolutional network is used to extract the deep features of the enzyme classification label information,and it is used to guide the multi-view learning process.Finally,the multi-label classifier was used to classify and predict the multifunctional enzymes.Compared with most of the existing classification and prediction methods of multifunctional enzymes based on deep learning,the proposed method has a certain improvement in the prediction performance of EC codes of each layer of multifunctional enzymes.(2)Although the above multifunctional enzyme classification prediction method ml DGCnet has achieved good prediction performance,it only learns the local features of the enzyme sequence when extracting features of the enzyme sequence.Since the function of a protein is usually closely related to its global structure,it is difficult to obtain effective global structural and functional information of an enzyme by only using local sequence features.In addition,the distribution of local features of enzyme sequences may be affected by the changes in enzyme structure,that is,the changes in enzyme structure may lead to different distribution of local features,making it difficult to accurately classify enzyme functions only relying on local features.To this end,based on the ml DGCnet method proposed in this paper,the CNN-Bi LSTM module used for sequence deep feature extraction in ml DGCnet is improved,and a multi-view deep learning multifunctional enzyme classification and prediction method ml CBi GCnet is proposed by fusing local and global sequence features.In the sequence feature extraction part of the method,a CNN-Bi GRUs hybrid network with multi-head attention mechanism is used to extract the deep local features and deep global features of the sequence,so as to better capture the overall structural information of multifunctional enzyme sequences.Thus,the prediction performance of the model is further improved to a certain extent.The experimental results show that compared with the ml DGCnet method in the last work,the prediction performance of ml CBi GCnet at each layer of EC encoding is greatly improved.