I have been working with Shannon's entropy (SE) and DNA lately, and referring to the formula and concept of SE, I just wondered whether one should in the case of DNA use each nucleotide as having a probability to show up as 0,25 or if it was better to use the probability based on the frequency in the genome (let's say the human genome).

In the case of the latter being the better, is it interesting to have those frequencies at the scale of the genome, the chromosome, the sub segment ?

I'm interested in calculating the entropy relative to a short window (the length of an Illumina read, about 125-150bases), using the proportion of each base on a given sequence length. The idea is "how much information is given" in this sequence, and by information I mean how much different bases do I have.

It might that I'm missing some basic mathematics knowledge here, don't hesitate to point it out so I can understand it better



1 Answer 1


I think that genome-wide probabilities (i.e. not 0.25 for each base) should be fine. For a given segment of DNA, the chance that the sequence doesn't appear anywhere else in the genome is fairly high.

There is lots of patchy non-randomness (e.g. at centromeres, STRs, and other tandem repeat regions), but there's not much point in modelling that because it's not random.

  • $\begingroup$ Thanks for your answer. So if I understand it well, not taking equivalent probabilities but the human genome bases probability is a better way ? $\endgroup$
    – Bratten
    Commented May 6, 2021 at 11:43
  • $\begingroup$ @Bratten Yes; I've made a slight update to my answer to clarify this. $\endgroup$
    – gringer
    Commented May 7, 2021 at 6:17

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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