Fat-tail test of regulatory DNA sequence

Abstract

Content-based computational methods for distinguishing between cis-regulatory element (CRE) and non-CRE are valuable for predicting CRE that have not been observed experimentally. The fluffy-tail test is one of the content-based CRE prediction methods. This is a bootstrapping procedure to identify abundant similar words with statistical significance in regulatory DNA, and then differentiate the regulatory DNA from non-CRE. The fluffy-tail test focuses only on the most frequently occurring subsequences in CREs, thus providing only a measure of homotypic transcription factor binding site (TFBS) clusters; however, most transcriptional regulatory regions contain multiple types of binding motifs; therefore, the fluffy-tail test may fail to capture statistical features arising from heterogeneous TFBS clusters in regulatory regions. In this paper, a kurtosis-based fat-tail test is proposed to measure both homogeneous and heterogeneous clustering. The results show that the fat-tail test separates CRE from non-CRE better than the fluffy-tail test.