Research On Prediction Method Of Protein-Protein Interaction Based On SVM And Protein Functional Annotation

The process of protein or protein formation is the process of protein-protein interaction(PPI),which is the result of the gradual formation of protein complexes by two or more protein molecule complexes.This study found that the current PPI data contained in the number of small protein can't meet the needs of the practical application of life sciences,such as common DIP and other PPI interaction database contains only more than 9000 human PPI interaction relationship,and commonly used genes The number of expression data in 1 to 2 million or so.Such as the commonly used differential expression analysis,found that many differentially expressed genes are not included in the known PPI,so a large number of protein interactions between the need to be predicted.The existing PPI data are obtained by means of experiments,including techniques such as tandem affinity purification and yeast two-hybrid technology.These experiments can achieve high accuracy but take too long and greatly reduce the cost of the experiment.Learning means to assist in predicting PPI interactions.Generally speaking,based on the machine learning algorithm to predict the interaction between the protein effect is quite good,but also has its own constraints,mainly reflected in: First,the prediction of PPI interaction between the application of machine learning algorithm with a supervisory role,training test PPI The relationship between the known and unknown PPI is relatively small,especially to determine the absence of interaction between the protein is less;Second,the vector feature that a single method,Or based on the PPI amino acid sequence method or based on gene co-expression methods,etc.,did not consider the PPI itself with other biological information;Third,the calculation is relatively large.In view of the above problems,the solution proposed in this paper:(1)The problem of SVM eigenvector representation This paper deals with the use of amino acid AC value,and also introduces the protein function annotation data of GO,KEGG and so on into the construction of eigenvector to construct a new eigenvector.(2)The PPI of the interaction based on the experiment is used as the training data set of the algorithm,and the negative experimental data set of the PPI with no interaction relation obtained by the current experiment or calculation means is found through the network search.Both positive and negative PPI dataset training tests SVM to predict PPI interactions.(3)Design and implement the algorithm based on computing method to help predict PPI-PPI_SPFA algorithm.In this paper,we propose a two-step strategy to filter PPI with little possibility of interaction,And then to predict the means.The PPI_SPFA algorithm has improved accuracy in predicting the interaction of PPI compared to other algorithms such as PPI_AC and i PPI-Esml.(4)In addition to the existing PPI interaction in the PPI interaction database,SVM predicts the remaining PPI interaction relationship,and finally constructs a relatively complete PPI interaction network.The future research focus is to combine SVM and protein function annotation GO,KEGG and so on PPI prediction algorithm to explore and innovate,improve PPI prediction algorithm accuracy and response speed.