Computational Approaches for the Prediction of Gene Regulatory Elements and the Analysis of their Evolutionary Conservation

Abstract

In the last several decades, the field of molecular biology has made substantial progress in deciphering the many facets of gene expression and regulation. The parallel development of experimental methods, especially high-throughput, greatly contributed to new possibilities for studying functional elements of the genome such as promoters and enhancers. In this thesis, I investigate the characterization and evolutionary conservation of enhancers. First, I present eHMM, a method that uses a supervised HMM with a constrained underlying Markov chain that incorporates prior biological knowledge about the molecular structure of enhancers in a dynamic model to predict heterogeneous enhancers of variable sizes on the basis of a minimal set of features. I demonstrate eHMM’s prediction performance using different validation setups within and across data sets, tissues and developmental stages and analyze genome-wide predictions in terms of functional genomic and epigenomic features, spatial accuracy, and susceptibility for false-positive results. Second, I investigate functional evolutionary conservation of enhancers in absence of detectable sequence conservation. For that, I introduce the concept of using multiple sets of pairwise alignments that allow moving through a species graph in order to produce accurate projections of non-alignable genomic regions between two species with large evolutionary distances. To that end, I present the methods IPP and SAPP that approach the task under slightly different aspects. IPP projects individual genomic point coordinates from one species onto another by interpolating their position between two alignable sequences, so-called anchor points. Instead of using only direct alignments between the two species in question, IPP implements the choice of an optimal set of bridging species that maximizes projection accuracy. I demonstrate IPP’s projection accuracy compared to using direct alignments, propose functional conservation to be a universal phenomenon, and identify individual occurrences of functional orthologs beyond sequence conservation. SAPP propagates anchor points rather than projecting genomic points in a fashion that minimizes resulting anchor spans in the target species. By that, it respects the conservation of synteny and provides maximally narrowed search spaces for analyzing enhancer equivalence between two species. Together, the work presented in this thesis aims at adding to our current understanding about the identity and the evolutionary properties of enhancers.