Effective Statistical Features for Coding and Non-coding DNA Sequence Classification for Yeast, C. elegans and Human

Abstract

This study performs a quantitative evaluation of the different coding features in terms of their information content for the classification of coding and non-coding regions for three species. Our study indicated that coding features that are effective for yeast or C. elegans are generally not very effective for human, which has a short average exon length. By performing a correlation analysis, we identified a subset of human coding features with high discriminative power, but complementary in their information content. For this subset, a classification accuracy of up to 90% was obtained using a simple kNN classifier.