Year: 2015
Type: dissertation
Abstract: This thesis focuses on the statistical analysis of Chromatin immunoprecipitation \nsequencing (ChIP-Seq) data produced by Next Generation Sequencing (NGS). ChIP-Seq \nis a method to investigate interactions between protein and DNA. Specifically, the method \naims to identify the binding sites of a particular protein of interest, such as a transcription \nfactor, in the genome. In the context of cancer research, this information is important \nto check whether, for example, a particular transcription factor can be considered as a \ntherapeutic target. \nThe sequence data produced by ChIP-Seq experiment are in the form of mapped short \nsequences, which are called reads. The reads are counted at each single genomic position, \nand the read counts are the data to be analysed. There are many problems related to the \nanalysis of ChIP-Seq data, and in this research we focus on three of them. \nFirst, in the analysis of ChIP-Seq data, the genome is not analysed in its entirety; instead \nthe intensity of read counts is estimated locally. Estimating the intensity of read counts \nusually involves dividing the genome into small regions (windows). If the window size \nis small, the noise level (low read counts) would dominate and many empty windows \nwould be observed. If the window size is large, the windows would have many small read \ncounts, which would smooth out some important features. The need exists for an approach \nthat enables researchers to choose an appropriate window size. To address this problem, \nan approach was developed to optimise the window size. The approach optimises the \nwindow size based on histogram construction. Note, the developed methodology is \npublished in [46]. \nSecond, different studies of ChIP-Seq can target different transcription factors and then \ngive different conclusions, which is expected. However, they are all ChIP-Seq datasets \nand many of them are performed on the same genome, for example the human genome. \nSo is there a pattern for the distribution of the counts? If the answer is yes, is the pattern common in all ChIP-Seq data? Answering this question can help in better understanding \nthe biology behind this experiment. We try to answer this question by investigating \nRUNX1/ETO ChIP-Seq data. We try to develop a statistical model that is able to describe \nthe data. We employ some observed features in ChIP-Seq data to improve the performance \nof the model. Although we obtained a model that is able to describe the RUNX1/ETO \ndata, the model does not provide a good statistical fit to the data. \nThird, it is biologically important to know what changes (if any) occur at the binding sites \nunder some biological conditions, for example in knock-out experiments. Changes in the \nbinding sites can be either in the location of the sites or in the characteristics of the sites \n(for example, the density of the read counts), or sometimes both. Current approaches for \ndifferential binding sites analysis suffer from major drawbacks. First, unclear underlying \nmodels as a result of dependencies between methods used, for example peak finding and \ntesting methods. Second, lack of accurate control of type-I error. Hence there is a need \nfor approach(es) to address these drawbacks. To address this problem, we developed three \nstatistical tests that are able to detect significantly differential regions between two ChIPSeq \ndatasets. The tests are evaluated and compared to some current methodologies by \nusing simulated and real ChIP-Seq datasets. The proposed tests exhibit more power as \nwell as accuracy compared to current methodologies.
Author: Ibrahim Nafisah
Language: English
Subfield: Molecular Biology
Domain: Life Sciences
Open Access status: green
Funder: King Saud University
OpenAlex ID: W2335970574