Transcription Factors influence the expression of target genes as either activators or repressors by binding to the DNA near genes. This binding is guided by TF sequence specificity. The closer the DNA is to the base preference, the more likely it is that the factor will bind. These motifs can be found both computationally and experimentally. There are three main approaches for discovering these motifs.
- Co-Regulation - In Lecture 11, we discussed a co-regulation type of discovery of motifs by finding sequences which are likely to have the motif bound. We can then use enumerative approaches or alignment methods to find these motifs in the upstream regions. We can apply similar techniques to experimental data where you know where motif is bound.
- Factor Centric - There are also factor centric methods for discovering motifs. These are mostly experimental methods which require a protein or antibody. Examples include SELEX, DIP-Chip, and PBMs. All of these methods are in vitro.
- Evolutionary - Instead of focusing on only one factor, evolutionary methods focus on all factors. We can begin by looking at a single factor and determining which properties we can exploit. There are certain sequences which are preferentially conserved (conservation islands). However, these are not always motifs and instead can be due to chance or non-motif conservation. We can then look at many regions, find more conserved motifs, and determine which ones are more conserved overall. By testing conservation in many regions across many genomes, we increase the power. These motifs have certain evolutionary signatures that help us to identify them: motifs are more conserved in intergenic regions than in coding regions, motifs are more likely to be upstream from a gene than downstream. This is a method for taking a known motif and testing if it is conserved.
We now want to find everything that is more conserved than expected. This can be done using a hill climbing approach. We begin by enumerating the motif seeds, which are typically in 3-gap-3 form. Then, each of these seeds is scored and ranked using a conservation ratio corrected for composition and small counts. These seeds are then expanded to fill unspecified bases around the seed using hill climbing. Through these methods, it is possible to arrive at the same, or very similar seeds in different manners. Thus, our final step consists of clustering the seeds using sequence similarity to remove redundancy.
A final method that we can use is recording the frequency with which one sequence is replaced by another in evolution. This produces clusters of k-mers that correspond to a single motif.
Validating Discovered Motifs
There are many ways that we can validate discovered motifs. Firstly, we expect them to match real motifs, which does happen significantly more often than with random motifs. However, this is not a perfect agreement, possibly due to the fact that many known motifs are not conserved and that known motifs are biased and may have missed real motifs. Positional bias. Biased towards TSS,
Motifs also have functional enrichments. If a specific TF is expressed in a tissue, then we expect the upstream region will have that factor’s motif. This also reveals modules of cooperating motifs. We also see that most motifs are avoided in ubiquitously expressed genes, so that they are not randomly turned on and off.
There are disadvantages to all of these approaches. Both TF and region-centric approaches are not comprehensive and are biased. TF centric approaches require a transcription factor or antibody, take lots of time and money, and also have computational challenges. De novo discovery using conservation is unbiased, but it can’t match motifs to factors and requires multiple genomes.