Using multi-label algorithm to predict the post-translation modification types of proteins

Abstract

Post-translational modifications (PTMs) play vital roles in most of the protein maturation, structural stabilization and function. How to predict protein' PTMs types is an important and challenging problem. Most of the existing approaches can only be used to recognize single-label PTMs type. By introducing the multi-labeled K-Nearest-Neighbor algorithm, a new predictor has been proposed which can be used to dispose of the proteins containing both single and multi-label PTMs type. As a result that the 10-fold crosses validation was implemented on a benchmark data set of proteins which were divided into the following 4 types: (1) methylation, (2) nitrosylation, (3) acetylation, (4) phosphorylation, where many proteins belong to two or more types. For such a complex system, the outcomes achieved by our predictor for the six indices were quite promising, anticipated the predictor may become a complementary tool in this area.