I am confused about the sequences one starts with to generate the PAM matrices.

The sequences to start with are supposed to diverge by no more than 1% (in sequence identity) wrt a common "inferred" ancestor. - if I understood correctly.

How is the common ancestor sequence inferred without having the means to do any sequence alignment to start with?

Does this mean that a PAM matrix is specific for an inferred ancestor, i.e. a protein family? - I got the impression there was only one in practice.

Thanks in advance,



1 Answer 1


Ideally, but I wouldn't worry because the chance of meeting the criteria in practice are slim. Most importantly you need a tree which is 'independently' derived from the protein sequence family in question. The PAM matrix is absolutely dependent on this. Generating a "true" tree without reference to the sequences you are studying isn't trivial, so it's got to generated with a different data set and the question is whether that data set gives a true tree, if you're stuck the DNA sequence of is a rough approximation (not ideal but you could justify it). If you can adequately do this, I wouldn't worry about the remaining assumptions because it's simply an optimisation which can't proven.

The PAM matrix generation will estimate the sequence of a common ancestor, based on a given tree. What this criteria is saying is that if you want a 'perfect matrix' the sequence reconstruction of the ancestral node needs 'x' because that is pivotal to calculating the matrix. If that's a bit out your matrix is a bit by a margin of error that isn't quantified (maybe it has been, see below)

Whether you can get get something close to the 'hub' sequence within your data set is one question and whether that sequence was included in the first place is all trial and error. I would assume the criteria are ascertained has been set by simulation studies (its fairly easy to do if you've got the time), which only affects the margin of confidence within the confines of that given simulation. PAM matrix generation has been around since the dawn of time and founding principle is an independent tree, everything else you can calculate and arm-wave, ie admit the calculation may not be so precise. In other it doesn't stop you doing it and presenting the results, that is perfectly valid, as long as simulated criteria are understood. The problem would be is if you ignore a past study and present your results 'without fault'. The biggest problem would be if you generated a tree and used the same data.to generate the matrix, because that is whatever matrix you used to generate the tree and that's a no no.

  • $\begingroup$ Alright, I see - so in practice, there is only one PAM1 matrix generated from all known protein sequences. (so to speak in the limit... I assume it converges to that) --- And when performing any pairwise alignment, I would compare my sequences to each other ensuring that no one pair has more than 1% divergence in identity - instead of comparing to that inferred common ancestor? $\endgroup$ Commented Jan 30, 2022 at 17:50
  • $\begingroup$ Whilst you have the gist, without supervision in coalescent theory this could be tricky. My real concern is the criteria are very theoretical and seldom met in real world data to be meaningful or practical. If If I was in your position I would justify my why I didn't meet the criteria.and move on. The tricky bit is our knowledge of theoretical basis of molecular phylogeny isn't identical and I don't really know your precise question or why it's been set. Given that I'd proceed with caution here, it could be a 'can of worms' $\endgroup$
    – M__
    Commented Jan 31, 2022 at 2:15

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.