Explore protein-protein interaction network involved in glucosinolate biosynthesis

Abstract

Protein is the primary element of organism and takes part in almost all the biological processes such as metabolism and neurological regulation. Generally, proteins are interacting with each other while they exert biological role in vivo. The exploration upon protein-protein interactions (PPIs) of the specific biological process could provide valuable information to the study of the relevant field. In this paper, we focus on the collection of proteins participated in glucosinolate biosynthesis, and build 4 decision tree models to predict PPIs involved in glucosinolate biosynthesis. Information of domain-domain interactions (DDIs) is introduced in constructing feature vectors, and the interactive or non-interactive relationship between two proteins is represented by a pair of symmetrical feature vectors. 4 domain-based decision tree models are constructed and trained by the samples with 1:1, 1:2, 1:3, 1:4 positive-negative ratio respectively. 5-fold cross-validation and a standalone external test are used in order to trace the best performed model. The proposed method is effective which is demonstrated by the higher specificity, sensitivity and high attribute usage while training decision trees. We use the intersection of the best two prediction results to validate and explore PPIs based on the proteins participated in glucosinolate biosynthesis, and finally a comprehensive PPI network is drawn according to the prediction result.