DNA sequence feature selection for intrinsic nucleosome positioning signals using AdaBoost

Abstract

Recent genome wide experiments indicate that DNA sequences themselves strongly influence nucleosome positioning as an intrinsic cell regulatory mechanism. While some sequence features are known to be nucleosome forming or nucleosome inhibiting, there is no systematic study on identifying optimal sequence features for quantitatively modeling of DNA binding affinity. In this paper, we propose a computationally efficient method of identifying a (small) number of sequence features for intrinsic nucleosome positioning. By using a modified version of AdaBoost, the proposed method is able to identify features to be used with a strong classifier to categorize nucleosome forming and nucleosome inhibiting local DNA sequences. Experimental results on extensive datasets show that the resulting classifiers give typically better prediction performance than the existing discrimination models on all the tested datasets with a much smaller number of features.