We can pretty easily quantify the amount difference between two different strings/sequences of characters. For example, if we take the words trebuchet and trebucket, we can say they have a Levenshtein distance of 1 (only one character-edit's worth of difference).

From a linguistic perspective, that distance is very noticeable, whereas the difference between compliment and complement is much less so. Even though both pairs are only one edit apart (respectively), not all differences are equally distinguishable.

What is the biological equivalent of this when it comes to DNA binding proteins and RNAs? How do I identify which DNA sequences are more recognizably distinct from each other? For example, if we took a DNA binding protein that recognizes the sequence TGCCTCGAA, is it more likely to misrecognize AGCCTGGAA than TGCCAGGAA (or vice versa) as its target sequence?

  • 2
    $\begingroup$ That's going to vary from one nucleic acid binding protein to another. Remember, DNA and RNA are not simply sequences of letters, but there can be secondary structure, as well as more basic chemical interactions between adjacent or more distant bases - see RNA hairpins as an example. All of this gives a "shape" to the recognized portion of nucleic acid, which can inform binding capacity just as much as, or potentially even more than primary sequence alone. $\endgroup$
    – MattDMo
    Apr 28 '16 at 20:24
  • $\begingroup$ @MattDMo That is certainly not surprising, but it would seem to me that there should be some parameter to that variation. For example, perhaps switching a single base in a sequence from a purine to pyrimidines is more distinguishable than to the other purine. $\endgroup$ Apr 28 '16 at 21:22
  • 3
    $\begingroup$ Related: biology.stackexchange.com/q/38808/3340 $\endgroup$
    Apr 29 '16 at 14:12

This isn't a question with a really well accepted answer yet, and comes up quite a lot in e.g. studies of population variation in transcription factor motifs.

Usually, we approximate the sequence preferences of a DNA-binding protein with a position weight matrix. A weight matrix will given you a score for two sequences, so the simplest means of quantifying the relative binding strengths for two sequences is compare these scores. You could also, say, compare the relative chances of getting the scores under some background distribution of scores, which would arguably be more comparable between different factors.

The PWM score (and other means of describing sequence specificity) are only an approximation to what you're really interested though, which is the binding energy associated with the interaction. If you have detailed experimental evidence about how the protein binds, you might use that, and there are some papers that build models for approximating it from PWM scores as well. However the reality is that you can only ever roughly approximate what's happening in the cell, because of molecular crowding, and the many other factors that will be present on the chromatin fibre, so the difference in PWM scores is often what gets used.

Predicting the impact of a given change to a non coding DNA sequence is a hard problem. Sometimes you'll get a decent approximation, but in the end you're trying to do biochemistry from first principles, and we aren't there yet.


DNA is a chemical, and therefore, its interactions are governed by its shape. There is no way to look at a DNA sequence and know all the ramifications that changing a letter will have on its shape. I could tell you that changing the first two or last two letters of an intron are highly likely to destroy a splice site, but you can't make hard and fast predictions about DNA binding in many other situations.


Not the answer you're looking for? Browse other questions tagged or ask your own question.