what is the problem with using clustal to do pairwise alignment?

Question

[glarue]

I have a set of putatively-paralogous protein sequence pairs that I'm trying to align. I (naively) used Clustal Omega to generate pairwise alignments, but am getting some weird downstream results (abnormally high dS values between many pairs) so I went back through the documentation and noticed that the Clustal suite insists that it is only to be used for multi-sequence alignments, not pairwise ones.

I am looking for some clarity as to what problems might be expected in using Clustal to generate pairwise alignments, so that I can better troubleshoot my data.

[Joe]

Did you eye-ball the alignment that you got? I also (naively perhaps) would have expected Clustal to be fine with pairwise too. You can check the alignment by eye though to see if it does look screwy.

[glarue]

So, I ran a comparison between Clustal Omega and NEEDLE (EMBOSS's recommended pairwise alignment tool) and, disconcertingly, with default options they both definitely give different alignments:

By eye, I guess the Clustal ones do seem worse but I'm not confident in my ability to assess them manually.

[Joe]

Were they using the same mismatch/gap/extend penalties?

[glarue]

Good point - no, they were both set to use the default values. I can't figure out how to set the gap penalties for Clustal Omega, but using ClustalW I tried setting the values to be the same as NEEDLE and this resulted in yet another, different, alignment.

[felixir]

Hi, I have recently came across the same problem as you and I can't seem to find the solution online. I wonder since you posted this question 4 years ago, have you managed to answer or find the solution to this problem? Thanks in advance

Answer 1

[Hussain Ather]

There are just differences between the methods of multiple sequence alignments and pairwise sequence alignments. Multiple sequence alignments can also use pairwise within their own methods. Clustal performs multiple sequence alignment by using pairwise sequence alignments and also constructing a guide tree for multiple alignment. It actually performs a "series of pairwise alignments of clusters of sequences." (Higgins et al 1988). For the alignment programs that only perform pairwise sequence alignments, dot-matrix methods, dynamic programming, and word methods might be used used.

[glarue]

Thanks for the response. What I'm after, though, is a better understanding of why EMBOSS recommends against using Clustal for pairwise alignments - presumably they're worse than those produced by, say, NEEDLE, but in what way? What sort of issues might I see in the alignments that would be ameliorated by using a dedicated pairwise alignment algorithm?