Identification of Active and Binding Sites with Multi-dimensional Feature Vectors and K-Nearest Neighbor Classification Algorithm

Abstract

Helicobacter pylori is a pathogenic and carcinogenic bacterium, mainly living in the stomach and duodenum, and has been declared a prokaryotic carcinogen by the World Health Organization. The control of gastric cancer has attracted increasing attention. Studying the binding reactions of substrates at different protein sites will help to understand the relationship between protein structure and function, and pave the way for future research on the pathogenesis of Helicobacter pylori and the development of protein-targeted drugs. This paper provides a new identification method for predicting protein sites. It wants to classify the active sites and binding sites of proteins based on a K-nearest neighbor classification method by learning the multi-dimensional features of protein sites. First, the protein information of Helicobacter pylori is retrieved, and the Active_site and Binding_site sites are obtained from the existing database. Then, the protein fragment sequences adjacent to the sites are intercepted, and the protein sequences are analyzed by a custom correlation function to obtain feature vectors with the same length. After that, supervised learning will be used. For the n-dimensional vector input after the transformation, the machine learning KNN classification algorithm is used to perform the corresponding kd-tree optimization, and the NCA algorithm is introduced to automatically learn the distance measurement and complete the dimensionality reduction. The accuracy rate in the test set reaches 84.2, which is 6.5 higher than the traditional gradient boosting tree algorithm (GBDT). It is shown that this classification method is much better than previous classifiers and can make the binding site of proteins more effective.