Computational representation and discovery of transcription factor binding sites
- Autores: Joan Maynou Fernández
- Directores de la Tesis: Alexandre Perera Lluna (dir. tes.)
- Lectura: En la Universitat Politècnica de Catalunya (UPC) ( España ) en 2016
- Idioma: inglés
- Tribunal Calificador de la Tesis: Beatriz Fabiola Giraldo Giraldo (presid.), José Manuel Soria Fernández (secret.), Agustín Gutiérrez Gálvez (voc.)
- Materias:
- Enlaces
- Tesis en acceso abierto en: TDX
- Resumen
The information about how, when, and where are produced the proteins has been one of the major challenge in molecular biology. The studies about the control of the gene expression are essential in order to have a better knowledge about the protein synthesis. The gene regulation is a highly controlled process that starts with the DNA transcription. This process operates at the gene level, hereditary basic units, which will be copied into primary ribonucleic acid (RNA). This first step is controlled by the binding of specific proteins, called as Transcription Factors (TF), with a sequence of the DNA (Deoxyribonucleic Acid) in the regulatory region of the gene. These DNA sequences are known as binding sites (BS). The binding sites motifs are usually very short (5 to 20 bp long) and highly degenerate. These sequences are expected to occur at random every few hundred base pairs. Besides, a TF can bind among different sites. Due to its highly variability, it is difficult to establish a consensus sequence. The study and identification binding sites is important to clarify the control of the gene expression. Due to the importance of identifying binding sites sequences, projects such as ENCODE (Encyclopedia of DNA elements), have dedicated efforts to map binding sites for large set of transcription factor to identify regulatory regions. In this thesis, we have approached the problem of the binding site detection from another angle. We have developed a set of toolkit for motif binding detection based on linear and non-linear models. First of all, we have been able to characterize binding sites using different approaches. The first one is based on the information that there is in each binding sites position. The second one is based on the covariance model of an aligned set of binding sites sequences. From these motif characterizations, we have proposed a new set of computational methods to detect binding sites. First, it was developed a new method based on parametric uncertainty measurement (Rényi entropy). This detection algorithm evaluates the variation on the total Rényi entropy of a set of sequences when a candidate sequence is assumed to be a true binding site belonging to the set. This method was found to perform especially well on transcription factors that the correlation among binding sites was null. The correlation among binding sites positions was considered through linear, Q-residuals, and non-linear models, alpha-Divergence and SIGMA. Q-residuals is a novel motif finding method which constructs a subspace based on the covariance of numerical DNA sequences. When the number of available sequences was small, The Q-residuals performance was significantly better and faster than all the others methodologies. Alpha-Divergence was based on the variation of the total parametric divergence in a set of aligned sequenced with binding evidence when a candidate sequence is added. Given an optimal q-value, the alpha-Divergence performance had a better behavior than the others methodologies in most of the studied transcription factor binding sites. And finally, a new computational tool, SIGMA, was developed as a trade-off between the good generalisation properties of pure entropy methods and the ability of position-dependency metrics to improve detection power. In approximately 70% of the cases considered, SIGMA exhibited better performance properties, at comparable levels of computational resources, than the methods which it was compared. This set of toolkits and the models for the detection of a set of transcription factor binding sites (TFBS) has been included in an R-package called MEET.
