The Prediction Of Protein Interaction Direction Based On Domain Physicochemical Properties

The direction of signal transmission between protein interactions is very important for most signal transduction in organisms.With the deepening of life science research,a large number of protein interaction data and the network of protein interactions composed of these data have been accumulated,and methods for annotating the protein interaction network by various forms have also appeared.However,only a few researchers have annotated the protein interaction network with signaling directions between protein interactions.In the current study,most methods for predicting the direction of signal transmission between protein interactions are based on protein networks.These methods usually do not utilize the annotated protein interaction data in the databases KEGG and NETPATH.A few protein domain-based methods are also based on protein interaction,constructing a feature matrix representing direction information based on the relationship between GO annotation or domain interaction and protein interaction,and then training the prediction model for classification.However,these methods do not take into account the inherent physical and chemical properties of the protein domain.In order to solve thisproblem,this study designed a method to predict the signal transmission direction between protein interactions based on ten physical and chemical properties of protein domains.Based on this method,a support vector machine model for prediction is trained.The model is optimized by selecting different kernel functions,penalty factors and kernel function parameters.Using the method of comparative experiment,the overall performance of the prediction model,the performance in different species and the physical and chemical factors affecting the accuracy of the prediction model were analyzed.The method based on the physical and chemical properties of the domain can be effectively used for the prediction of signal transmission direction between protein interactions,and this method is extended to the prediction of protein interaction categories(activation/inhibition).First,with the protein interaction data with specific directions and the relationship between domain interaction and protein interaction,we calculate ten physical and chemical properties of protein domains,and use them to construct feature matrix that can represent the information of protein interaction direction.A support vector machine model for signal direction prediction between protein interactions is trained.The kernel function,penalty factor and kernel function parameters of the support vector machine model are selected by comparison experiment and the grid search algorithm.The prediction model is completed optimized.Then,through five-fold cross-validation,the overall performance of theprediction model,the performance on different species datasets,and the physical and chemical properties of the domain that affect the accuracy of the prediction model are analyzed and compared?In contrast,it is found that the prediction models based on ten physical and chemical properties of the domain can effectively and stably predict the signal transmission direction between protein interactions,and the prediction results on the data sets of highly evolved species are highly reliable.With the method based on the physicochemical properties of the domain,we attempted to predict the activation/inhibition categories between human protein interactions.It can provide a reference for signal direction and class prediction between protein interactions.In general,this paper proposes a method for signal transmission direction between protein interactions based on the prediction of physicochemical properties of the domain.The support vector machine prediction model established by this method can be effectively used for signal transmission between protein interactions.In addition,this study also explores the physicochemical properties of the domain that affect the accuracy of the prediction model,and proves the necessity of ten physical and chemical properties of the domain.A domain-based approach was extended to the study of human protein interaction classes for simple calculations.The results provides new insights for researchers to further annotate protein interaction networks using protein interaction directions and classes.