Incorporating evolution of transcription factor binding sites into annotated alignments

Abstract

Identifying transcription factor binding sites (TFBSs) is essential to elucidate putative regulatory mechanisms. A common strategy is to combine cross-species conservation with single sequence TFBS annotation to yield "conserved TFBSs". Most current methods in this field adopt a multi-step approach that segregates the two aspects. Again, it is widely accepted that the evolutionary dynamics of binding sites differ from those of the surrounding sequence. Hence, it is desirable to have an approach that explicitly takes this factor into account. Although a plethora of approaches have been proposed for the prediction of conserved TFBSs, very few explicitly model TFBS evolutionary properties, while additionally being multi-step. Recently, we introduced a novel approach to simultaneously align and annotate conserved TFBSs in a pair of sequences. Building upon the standard Smith-Waterman algorithm for local alignments, SimAnn introduces additional states for profiles to output extended alignments or annotated alignments. That is, alignments with parts annotated as gaplessly aligned TFBSs (pair-profile hits)are generated. Moreover,the pair- profile related parameters are derived in a sound statistical framework. In this article, we extend this approach to explicitly incorporate evolution of binding sites in the SimAnn framework. We demonstrate the extension in the theoretical derivations through two position-specific evolutionary models, previously used for modelling TFBS evolution. In a simulated setting, we provide a proof of concept that the approach works given the underlying assumptions,as compared to the original work. Finally, using a real dataset of experimentally verified binding sites in human-mouse sequence pairs,we compare the new approach (eSimAnn) to an existing multi-step tool that also considers TFBS evolution. Although it is widely accepted that binding sites evolve differently from the surrounding sequences, most comparative TFBS identification methods do not explicitly consider this.Additionally, prediction of conserved binding sites is carried out in a multi-step approach that segregates alignment from TFBS annotation. In this paper, we demonstrate how the simultaneous alignment and annotation approach of SimAnn can be further extended to incorporate TFBS evolutionary relationships. We study how alignments and binding site predictions interplay at varying evolutionary distances and for various profile qualities.

Evolution of gene regulation of pluripotency - the case for wiki tracks at genome browsers

“…"Assuming that conservation goes with functional importance, "phylogenetic profiling"‚ and related methods [13-22]. .."…”

“…Many transcription factor binding site prediction tools exploit libraries of known binding motifs and evolutionary conservation, and usually they infer sets of related sites (cis-regulatory modules). Assuming that conservation goes with functional importance, "phylogenetic profiling" and related methods [13-22] suggest that predicted binding sites are the more likely to be functional, the more conserved they are. The integrated analysis of the evolution of cis-regulatory modules and the network of regulators is in its infancy.…”

Position-specific evolution in transcription factor binding sites, and a fast likelihood calculation for the F81 model

“…Bais et al . [23] considered TFBS evolution in the context of annotated genome alignments; they consider both the F81 model and the Halpern-Bruno model. Mustonen et al .…”

Tools for simulating evolution of aligned genomic regions with integrated parameter estimation

“…A few concrete examples of applications that have been critically evaluated using simulated data are protein-coding genefinding [ 1 ], identification of transcription factor binding sites [ 2 ] and other functional non-coding DNA [ 3 ], phylogenetic reconstruction using distances based on substitutions [ 4 , 5 ] or indels [ 6 ], protein homology detection [ 7 ] and protein multiple alignment [ 8 ]. This list is a gross under-sampling from recent years, based on a citation search using the simulators described below; since simulation is almost a prerequisite for any new tool, the full list is much longer than this.…”