Behavior and Limitations of Motif Finding

Today, we’ll look at some behaviors of the popular MEME motif-finding software. MEME is a powerful tool for discovering putative regulatory motifs in DNA sequences. You can find it at the MEME Suite website.

To begin, we’ll focus on a motif-finding problem in D. melanogaster promoter sequences. The sequences we will work with are in the FASTA-formatted file dmel-promoters.fna. These sequences were extracted from just upstream of the transcription start sites of twelve genes, all of which are found on the dot chromosome and exhibit expression specific to fly heads. We hypothesize that these genes, which share both common expression and a common chromosomal environment, are regulated by one or more shared transcription factors.

To investigate this hypothesis, we’ll use MEME to look for putative motifs shared among these sequences. In the submission form at the above URL, click on the "Browse" or the "Choose File" button under the "Input the primary sequences" field to tell MEME that you want to search dmel-promoters.fna. Finally, hit "Start Search" to launch the motif finder.

You might have to wait a few minutes for the search to finish, depending on the server load. The search page will refresh periodically. When the search is done, new links will appear on the "Results" section of the page. Click on the "MEME HTML output" link to view the search result.

MEME returns several putative motifs, each with a very significant E-value. These motifs vary between 41 and 50 bases in length. You can click on the down arrow under the "More" heading next to each motif to see a detailed alignment of its instances.

Unfortunately, we can’t tell much about a putative motif’s function just by looking at the raw MEME output. However, just as BLAST lets us compare new sequences to a database of well-annotated functional elements, MEME has a companion tool, Tomtom, that can compare motifs to known transcription factor binding sites.

Let’s submit the first motif in the output to Tomtom. Click the right-facing arrow under "Submit/Download" next to the first motif. The "Tomtom" option should already be selected in the resulting dialog. Hit "Submit" to pull up the Tomtom submission page for your motif.

Just like BLAST, Tomtom offers a choice of organism-specific databases. Under "Select target motifs," specify "FLY (Drosophila melanogaster) DNA" in the first drop-down menu. The second menu now offers several motif databases to search. Select the "Fly Factor Survey" database, and hit "Start Search" to continue. Once the search is complete, click on the "Tomtom HTML output" link to view the search result.

For each motif from Fly Factor Survey that matches part of our putative motif, Tomtom gives an estimate (in particular, an E-value) of how significant the match is, and a link to more information about the gene whose protein product is associated with the motif. Among other things, these links provide information about the functional roles associated with the protein.

The quality and quantity of MEME's output is quite sensitive to its search parameters. Let’s try a couple of variants of our original search to see how these parameters affect the results we obtain.

Go back to the MEME search page. Repeat the MEME search, but this time, open the "Advanced options" tab and limit the width of the motifs returned to at most 20, rather than the default of 50.

Another important parameter for MEME is how many times a motif is permitted to occur in a sequence. By default, MEME assumes that a motif occurs either once or not at all in each input sequence. However, we can instead tell MEME that a motif could occur any number of times in each sequence.

To see if a motif might occur multiple times, look at the "Motif Locations" tab in the MEME results. By default, it shows the (at most one) instance in each sequence that contributed to MEME's reported motif model. If we instead select the "Motif Sites+Scanned Sites" option, it will also show (in lighter colors) other parts of the sequences that might match one of the motifs but were not used to build the motif model. Try this now.

If a motif occurs in multiple instances per sequence, it might be a good idea to consider all these instances when determining the motif’s specificity in the first place. To do this, rerun the MEME search, still limiting the maximum motif length to 20, but this time, choose the "Any Number of Repetitions (anr)" option under the "Select the site distribution" section.

For our last exercise, we’ll apply MEME to what should be quite an easy problem. The file planted-100.fna contains 20 sequences, each of length 100 bases (much shorter than the previous examples). These sequences were generated uniformly at random with equal base frequencies. Into each sequence, we planted an instance of a single motif of length 15. The motif has consensus sequence TTTTGTCGCATTCCG, but each instance differs from this consensus by four substitutions at randomly chosen positions. All instances were planted on the forward strand.

This is a very strong motif, in the sense that a set of unrelated 15-mer sequences with this degree of similarity would be highly unlikely to occur by chance in the background sequence; MEME would therefore give it a significant E-value. You can see the actual starting coordinate of each instance in its sequence in the sequences' FASTA headers.

Ask MEME to find the planted motif in this file. To make MEME's job easier, change the search parameters to tell it that there is exactly one motif instance per sequence, that the minimum and maximum widths are both 15, and that it should search only the forward (given) strand.

The file planted-600.fna contains another planted motif of length 15, this time with consensus TCACTATATCAGTCT, again with four substitutions in each instance. This time, however, the 20 random background sequences are each 600 bases long. Despite the somewhat longer background length, the planted motif remains highly significant. Try to find this motif with MEME, using the same parameters as before.