Nucleic Acids Research Advance Access originally published online on December 23, 2008
Nucleic Acids Research 2009 37(3):939-944; doi:10.1093/nar/gkn1019
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Nucleic Acids Research, 2009, Vol. 37, No. 3 939-944
© 2008 The Author(s)
This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/2.0/uk/) which permits unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited.
Computational Biology |
Pseudocounts for transcription factor binding sites
1Department of Medical Genome Sciences, Graduate School of Frontier Sciences, University of Tokyo, 5-1-5 Kashiwanoha, Kashiwa, Chiba 277-8562, 2Computational Biology Research Center, Institute for Advanced Industrial Science and Technology, Tokyo 135-0064, 3Human Genome Center, Institute of Medical Science, University of Tokyo, 4-6-1 Shirokanedai, Minato-ku, Tokyo 108-8639 and 4Institute for Bioinformatics Research and Development (BIRD), Japan Science and Technology Agency (JST), 5-3 Yonbancho, Chiyoda-ku, Tokyo 1002-0081, Japan
*To whom correspondence should be addressed. Tel: +81 3 5449 5131; Fax: +81 3 5449 5133; Email: knakai{at}ims.u-tokyo.ac.jp
Received September 24, 2008. Revised November 12, 2008. Accepted December 5, 2008.
To represent the sequence specificity of transcription factors, the position weight matrix (PWM) is widely used. In most cases, each element is defined as a log likelihood ratio of a base appearing at a certain position, which is estimated from a finite number of known binding sites. To avoid bias due to this small sample size, a certain numeric value, called a pseudocount, is usually allocated for each position, and its fraction according to the background base composition is added to each element. So far, there has been no consensus on the optimal pseudocount value. In this study, we simulated the sampling process by artificially generating binding sites based on observed nucleotide frequencies in a public PWM database, and then the generated matrix with an added pseudocount value was compared to the original frequency matrix using various measures. Although the results were somewhat different between measures, in many cases, we could find an optimal pseudocount value for each matrix. These optimal values are independent of the sample size and are clearly correlated with the entropy of the original matrices, meaning that larger pseudocount vales are preferable for less conserved binding sites. As a simple representative, we suggest the value of 0.8 for practical uses.