< Terug naar vorige pagina
Developing algorithms for the in silico identification of transcription factor binding sites
Boek - Dissertatie
Modeling the specificity of transcription factors to the DNA is one of the challenges that has kept many bioinformatics researchers busy since the early beginnings. Initially it was expected that a universal recognition code describing the amino acid to base pair contacts would be able to describe protein-DNA complex formation. However, until this very day a universal recognition code has not yet been found and alternative methods became more important. Nowadays, methods that describe the specificity of only one transcription factor (or a small family of transcription factors) are used most often. These methods make use of a set of experimentally validated binding sites to construct a profile for each transcription factor. One of the oldest profile-based methods is the consensus sequence method. Consensus sequences consist of a simple text string in which each character of the string represents the most prevalent nucleotide in the corresponding position of DNA binding sites. As an extension to these consensus sequences, in 1982, Gary Stormo introduced the well-known and very popular positional weight matrix (PWM). These PWMs consist of a 4xL matrix, with L being the length of the binding sites. In each row of these matrices, the frequency of occurrence of one of the four nucleotides is given for a certain position in the binding sites. Even though these PWMs are a big improvement to the consensus sequences method, they also lead to many false positive predictions. Many alternative methods try to improve the accuracy of these PWMs, most of them with very limited success. In this thesis I will discuss the shortcomings of the previous generation of prediction methods and I will suggest new methods that overcome some of these shortcomings.The first method that will be discussed in this thesis makes use of a multiple sequence alignment (MSA) to visualize evolutionary conserved transcription factor binding sites that are predicted with the PWM method. Binding sites that are conserved across all species in these alignments have a higher likelihood to be functional. Mutation of these binding sites would result in a less fit species, therefore mutations in these binding sites would have a negative effect. By inspecting these multiple sequence alignments for putative PWM hits we can reduce a large number of false positive predictions as false positive hits are less likely to be conserved.A second contribution of this thesis to the improvement of prediction methods is the research on and development of a number of new methods that make use of the structure and the biophysical characteristics of protein-DNA complexes. These characteristics are often overlooked in the previous generation of prediction methods even though they are very important for binding specificity in many protein-DNA complexes. With the help of the Random Forest classification method and sequence-based structural and biophysical characteristics we managed to develop models that can predict transcription factor binding sites with a higher level of accuracy. Based on this method, we also developed a user-friendly web-tool that can make use of a large number of pre-calculated transcription factor models.
Pagina's: XXII, 149 p.
Jaar van publicatie:2014