Back to search

Freie Universität Berlin

Statistics for transcription factor binding sites

Abstract

dc:description.abstract

Transcription factors (TFs) play a key role in gene regulation. They interact with specific binding sites or motifs on the DNA sequence and regulate expression of genes downstream of these binding sites. In silico prediction of potential binding of a TF to a binding site is an important task in computational biology. From a statistical point of view, the DNA sequence is a long text consisting of four different letters ('A','C','G', and 'T'). The binding of a TF to the sequence corresponds to the occurence of a word in the sequence, e.g. 'AACCTC'. Hence, word count statistics can be applied to problems such as number of binding sites and distances between binding sites. The major problem in word count statistics are dependencies between sequence positions. These dependencies arise due to possible overlaps of words. So far, exact formulae to compute the count distribution of clustered occurrences only exist based on generating functions. We newly derive a recursive formula and use it to obtain a normal approximation. In fact, a TF does not bind to one single word but allows mismatches and substitutions. This is captured in a statistical model called Position Frequency Matrix (PFM). A PFM assigns a score to each position of the word and letter. If the summed score reaches a certain threshold, the TF is assumed to bind to that sequence region. In fact, one can transform this representation to a set of words which are bound by the TF. Unfortunately, enumeration of the set of words takes exponential costs. In addition, the set of words grows enourmously for longer binding sites (around 500,000 for a binding site of length 15). Hence, word count statistics and its approximations become inefficient and very inaccurate. Therefore, the need for new statistics and efficient algorithms arises. Instead of enumerating all words, we use a statistical representation - the PFM - and model dependencies explicitly. In fact, probabilities for overlaps are dependencies of the summed scores between two positions. Hence, we reduce the problem to computing the two dimensional convolution of the score distributions for each possible overlap and derive an exact formula for the variance of PFM counts. Furthermore, we found an accurate approximation for the distribution of the number of occurrences using a compound Poisson distribution. Our approximation outperforms all alternative approaches. In addition, we give Poisson statistics for the number of occurrences without overlaps such that other standard word count statistics (like distances between occurences) can be applied. Third, we develop statistics to compute the significance of co- occurrences and co-operativity among sets of TFs. Fourth, we use the variance to define a natural measure of similarity between DNA motifs. We explicitly state formulae for PFMs. Compared to standard approaches, it shows higher correlation with empirical data. It also allows to cluster sets of TFs and gives results comparable with more sophisticated clustering algorithms. Finally, we use this similarity measure to compute the representation quality of PFMs for a set of experimentally verified binding sites. Besides a threshold optimization method which significantly improves the quality of PFMs in Transfac and Jaspar, we can indeed select DNA motifs, which violate PFM assumptions and, therefore, cannot be reasonbly represented as PFMs.

Degree

thesis:*
Department dc:contributor.department
Mathematik und Informatik
Year dc:date.issued
2008

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Pape, Utz J.
Committee members dc:contributor.committeemember
  • Prof. Dr. Martin Vingron
  • Prof. Dr. Sven Rahmann
  • Prof. Dr. Dr. h.c. Peter Deuflhard

Subjects

dc:subject × 9

Rights

dc:rights.accessrights
Statement dc:rights.accessrights
  • open access
Language dc:language
eng

Identifiers

dc:identifier.*
URN
urn:nbn:de:kobv:188-fudissthesis000000005801-7
Dc Identifier Other
FUDISS_thesis_000000005801
FUDISS_derivate_000000004570

Chain of custody

source
Harvested from
Freie Universität Berlin
Base URL
refubium.fu-berlin.de/oai/request
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
citation

Pape, Utz J.. Statistics for transcription factor binding sites. 2008. https://refubium.fu-berlin.de/handle/fub188/4591