Research On Prediction Approach Of Protein-Protein Interactions

Protein-protein interactions (PPIs) which are pivotal for understanding the principle of life, take part in multiple kinds of life processes. The data of biochemical experiment demonstrated PPIs is accumulating and the PPIs prediction method also renovating with the developing of science and technology. Thus, it has both theoretical and practical significance to research in PPIs prediction.In this paper, domain information and domain-domain interactions (DDIs) information are applied together to building the feature vectors for training decision tree (DT) PPIs predictor and support vector machine (SVM) PPIs predictor. A couple of pairing vectors is applied to represent the relationship between two protein and training the DT-PPIs predictor. DT-PPIs predictor is used to refine the not experiment demonstrated false positive PPIs in arabidopsis PPI database (AtPIN),in order to achieve this, the5-fold cross-validation of multiple ratio negative training sample and positive training sample are used to detect the sensitivity, and a standalone negative test sample is used to detect the specificity. In accord with the D-DDIs feature vectors, a D-DDIs kernel is proposed and verified while training the SVM-PPIs predictor. Multiple density of grid search are used for optimizing parameters during building the SVM-PPIs predictor. According to the gene product annotated by gene ontology (GO) of biosynthesis process of glucosinolate which is secondary metabolism product in arabidopsis and the biology articles of genes participate in it, a relatively credible and integrated PPI network of glucosinolate pathway is formed by using the DT-PPIs predictor to refine and filter out the false positive in AtPIN. The SVM-PPIs predictor is applied for predicting all the potential PPIs of the gene AT1G74090or AT5G07690either of which does not participate in PPI network. Furthermore, the division of union and intersection and longest similar path are used to calculate the similarity of PPIs base on the cell component GO in order to classify the result of prediction.