Recognition of beta-alpha-beta motifs in proteins by using Random Forest algorithm

Abstract

A beta-alpha-beta motif was dataset constructed by using the Definition of Secondary Structure of Proteins (DSSP) and PROMOTIF software, that analyzes a protein coordinate file and provides details about the structural motifs in the protein. We performed a statistical analysis on beta-alpha-beta motifs and non-beta-alpha-beta motifs, and the study objects that loop-helix-loop length was from 10 to 26 amino acids were selected. Hydropathy component of position and amino acid composition were combined as characteristic parameter for expressing the sequence characteristics. A Random Forest algorithm for predicting beta-alpha-beta motifs was developed. The overall accuracy and Matthew's correlation coefficient of 5-fold cross-validation achieved 88.9% and 0.78.