Computational methods for dissecting transcription regulatory networks

Klishami, Saeed Omidi. Computational methods for dissecting transcription regulatory networks. 2015, Doctoral Thesis, University of Basel, Faculty of Science.

Available under License CC BY-NC (Attribution-NonCommercial).


Official URL: http://edoc.unibas.ch/diss/DissB_11328

Downloads: Statistics Overview


Gene expression is regulated on many levels of which transcription regulation is the
most studied. Transcription factors (TF) arguably are the most integral part of transcriptional
regulation; by recognizing and specifically binding to short, but degenerate,
DNA sequences, TF control the transcription initiation of genes. In the mid 70s, the first
mechanism of TF binding recognition was introduced which is based on simple hydrogen
bonding rules. Since then numerous observations of complex TF binding site recognition
has substantially altered our viewpoint of the TF binding specificity. In particular, recent
studies have uncovered subtle pairwise dependencies (PD) between positions within
binding sites, which disapproves the common assumption in many current computational
models{ that binding positions contribute independently toward the binding affinity of
a sequence. Several works already tried to incorporate PD within a framework of binding
site recognition, but due to the complexity of this problem, they failed to provide
a consistent and rigorous methodology. In my PhD, we have addressed PD from a
computational perspective by introducing dinucleotide weight tensors (DWT), which
incorporates the entire information on PD into a robust mathematical model. Among
several advantages, the DWT model does not have any tunable parameters which makes
it highly applicable. Finally, recent boom of high-throughput data has provided a unique
window to investigate various questions regarding to binding specificity. Here, we have
systematically tested the DWT model against the classical non-dependent model over a
large number of human TF ChIP-seq data. The in vivo data has clearly demonstrated
the role of PD in binding specificity. Remarkably, we found resulting dependency models
from ChIP-seq data, outperform non-dependent models on separate in vitro data. In
fact, testing over the HT-SELEX data, a high-throughput variant of the SELEX, has
further corroborated the importance of PD.
Advisors:Nimwegen, Erik van
Committee Members:Bergmann, Sven
Faculties and Departments:05 Faculty of Science > Departement Biozentrum > Computational & Systems Biology > Bioinformatics (van Nimwegen)
Item Type:Thesis
Thesis Subtype:Doctoral Thesis
Thesis no:11328
Thesis status:Complete
Number of Pages:76 Bl.
Identification Number:
edoc DOI:
Last Modified:08 Feb 2020 14:05
Deposited On:26 Aug 2015 14:31

Repository Staff Only: item control page