0
$\begingroup$

This is probably a naive question.

I have accesssed the Clustal Omega online multiple sequence alignment tool at http://www.ebi.ac.uk/Tools/msa/clustalo/ for the first time, but I think I need to provide more input to get it to do what I want, by specifying more constraints.

Edited to clarify constraints:

The original sequence P sequence has four symbols, say A,B,C,D.

The algorithm has access to 4 derived sequences.

P1: all occurrences of A,B deleted, no other change

P2: all occurrences of A,D deleted, no other change

P3: all occurrences of C,D deleted, no other change

P4: all occurrences of B,C deleted, no other change

Goal: given P1,P2,P3,P4, recover P

Example

P: ABACCADDACBBC

P1 CCDDCC

P2 BCCCBBC

P3 ABAAABB

P4 AAADDA

Note that I have removed the under bars by which I had tried to indicate deleted symbols but since the deletion processes for the Pi are deterministic they can be supplied to Clustal or whichever tool is appropriate.

End of edit

Toy madeup example illustrating my input, the output, and the problem.

The actual sequence from which these sequences came after deletions, followed by the resulting sequences is as below:

p ABACBAC

p1 A_AC_A_

p2 _BA__AC

p3 __ACBA_

so I input

>p1
AACA
>p2
BAAC
>p3
ACBA

and got the output

p1 _AACA_
p2 BAAC__
p3 __ACBA
      ** 

Which is entirely reasonable, but I have constraints.

I want to align with the constraints that

  • for certain symbols matches between certain sequence pairs should be maximized, and
  • for certain symbols mismatches between certain sequence pairs are allowed

In fact given p1,p2,p3,p4, which all arose from p after deletions, there are the pairwise constraints

  • A should be a match between p3 and p4 (it never appears in p1 or p2)
  • B should be a match between p2 and p3 (it never appears in p1 or p4)
  • C should be a match between p1 and p2 (it never appears in p3 or p4)
  • D should be a match between p1 and p4 (it never appears in p2 or p3)

This means that p1 has only C,D; p2 has only B,C; p3 has only A,B; and p4 has only A,D characters after deletion.

I want to obtain the most likely alignment or a few most likely alignments of p1,p2,p3,p4 under these constraints.

I have looked to see if I could figure out how to input the constraints but no luck.

Should I have used another tool? In general I will have much longer sequences, typically 4 of them, with pairwise agree/disagree constraints, as stated above.

$\endgroup$
  • 1
    $\begingroup$ This question is unclear. What do you mean by A,B, C and D? Are these residues, or points in a list. I think you should expand on your toy sequences.Assuming you can't post your actual sequences, perhaps use another real peptide and make up similar constraints to the ones in your own problem (although I still don't understand what your constraints are). $\endgroup$ – James Apr 17 '16 at 2:59
  • 1
    $\begingroup$ I also don't understand your constraints. Also, why do you need constraints? Usually when you do MSA, you don't pose any constraints since you presumably don't know how the query sequences can be aligned. Also, I bet that if you include sequence "p" in your MSA, the alignment will be different and will include your constraints. $\endgroup$ – Gergana Vandova Apr 17 '16 at 5:10
  • $\begingroup$ @James, thanks for your comments, I will re-edit the question to clarify. $\endgroup$ – kodlu Apr 17 '16 at 22:04
  • $\begingroup$ @GerganaVandova, thanks, will rephrase the question. $\endgroup$ – kodlu Apr 17 '16 at 22:05
3
$\begingroup$

Answer

You can't. Clustal doesn't provide the options of setting the sort of restraints you would like. Multiple Sequence Alignment (MSA) is difficult to program and the authors of Clustal have been refining their algorithm for years. If it were perfect and completely robust they might be in a position to add more user tweaking, but at the moment you're stuck with those provided.

Comparison matrices and forcing alignments

It may be worth while to mention how one can force particular correspondences in simpler pairwise alignments. This involves manipulation of the comparison matrix used to score the comparison. DNA comparison matrices may be very simple, scoring, say, 1 for a perfect AA, TT, GG, CC match and 0 for any mismatch. Protein comparison matrices are more complex, with the values for perfect matches dependent on rarity and unique function (e.g. In the PAM250 substitution matrix WW scores 17, whereas GG scores only 5) and the values for mismatches also varying markedly (e.g. ED scores 4, compared with 5 for EE; but EW scores –7). There are various matrices available, so implementations of the Smith Waterman and Needleman and Wunsch pairwise comparison algorithms generally allow you to specify your own.

Suppose you have biological information (not intrinsic in the algorithm) that makes you wish to force an alignment. For example, AAATAA sequences near the polyA tail of two mRNA 3'UTRs should align, but sometimes do not because other stretches of As give better scores. What you can do is replace the AAATAA by AAAZAA in each case, and then modify the scoring matrix so that ZZ scores very highly (e.g. 20) compared with AA. This will force the alignment, after which you can change the Zs back to Ts.

Multiple Sequence Alignments tend to to use position-specific dynamic scoring matrices. As the overall sequences align, the original scoring system is adjusted to reflect the frequency found at a particular position. (For example if a position is skewed to E in 90% if cases, EE will score more than at a position in which E is found only in 20% of cases).

I'm not saying it would not be possible to change the initial Clustal matrix with Zs at a particular position, but I don't think this possibility is provided. (Anyway your situation is more complex.) Glancing at the Clustal Omega paper, I see that they allow, indeed encourage, you to build up alignments from previous profiles. If you managed to force the alignments of a few sequences and input the profile derived from that, you might get what you want. But I imagine that requires the command-line program from Dublin, rather than the EBI online version.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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