Research Article

A discriminative method for protein remote homology detection based on N-Gram

Published: January 15, 2015
Genet. Mol. Res. 14 (1) : 69-78 DOI: https://doi.org/10.4238/2015.January.15.9
Cite this Article:
S. Xie, P. Li, Y. Jiang, Y. Zhao (2015). A discriminative method for protein remote homology detection based on N-Gram. Genet. Mol. Res. 14(1): 69-78. https://doi.org/10.4238/2015.January.15.9
2,573 views

Abstract

Protein remote homology detection refers to detecting structural homology in proteins with an extremely low rate of sequence similarity. Such detection is primarily conducted using 3 methods: pairwise sequence comparisons, generative models for protein families, and discriminative classifiers. In this study, a discriminative classification method involving N-Grams was adopted to extract features using a random forest algorithm to classify data sets. Experiments in the SCOP 1.53 data set showed that our approach improved the receiver operating characteristic by 6% compared with well-known methods. To determine a score threshold that could be used to divide the data set, we also used a heuristic method through which the precision of positive examples and recall rate reached 0.5647 and 0.8647, respectively. Few other studies have investigated the recall and precision of such examples.

Protein remote homology detection refers to detecting structural homology in proteins with an extremely low rate of sequence similarity. Such detection is primarily conducted using 3 methods: pairwise sequence comparisons, generative models for protein families, and discriminative classifiers. In this study, a discriminative classification method involving N-Grams was adopted to extract features using a random forest algorithm to classify data sets. Experiments in the SCOP 1.53 data set showed that our approach improved the receiver operating characteristic by 6% compared with well-known methods. To determine a score threshold that could be used to divide the data set, we also used a heuristic method through which the precision of positive examples and recall rate reached 0.5647 and 0.8647, respectively. Few other studies have investigated the recall and precision of such examples.

About the Authors