Python Programming for Biology: Bioinformatics and Beyond

13

Multiple-sequence alignments

Contents

Multiple alignments

Progressive pairing

Alignment consensus and profiles

Generating a consensus sequence

Generating an alignment profile

Profile alignments

Generating simple multiple alignments in Python

Profile-based multiple alignment

Interfacing multiple-alignment programs

Using ClustalW from Python

Multiple alignments

Expanding  from  an  alignment  of  just  two  sequences,  the  more  sequences  you  can  align

together  then  the  more  information  you  have  and  the  more  accurate  your  alignment  will

be.  The  caveat  to  this  is  that  the  closeness  of  the  relationship  between  sequences  is

important and should also be taken into account. Consider that for two very closely related

sequences,  the  differences  are  significant  and  similarities  less  significant,  because  you

expect  similarity.  In  contrast,  for  distantly  related  sequences  residue  differences  are

positively expected, so the similarities are more significant and differences less so.

Given that we have described how alignments can be made for pairs of sequences, the

next topic is to show how we can include more than just two sequences in an alignment to

make a multiple-sequence alignment, and how the overall or average properties of such an

alignment  can  be  measured.  As  a  naïve  example  of  the  benefits  of  multiple-sequence

alignment, consider the following alternative alignments for two sequences:

GCGCATG--GCGCAT GCGCAT--GGCGCAT

GGGCATGCGGCGCAT GGGCATGCGGCGCAT

There  is  no  way  to  know  which  alignment  is  best;  the  gap  appears  equally  good  in

either position. However, if there is a third sequence which supports one scenario over the

other then we can make a better judgement, in this instance supporting the first scenario.

GCGCATG--GCGCAT

GGGCATGCGGCGCAT

GCGCATGCCCCGCAT

---

When  aligning  pairs  of  sequences,  which  you  can  imagine  as  a  two-dimensional

problem,  we  can  use  the  dynamic  programming  trick.  However,  as  the  number  of

sequences  in  a  multiple  alignment  increases  the  complexity  of  the  problem  increases;  in

effect we get a new dimension of possibilities for each extra sequence. If the comparison

of two sequences required a grid of points then three requires a volume of points, four a

4D hypervolume etc. In essence the alignment problem grows very complex very quickly

with extra sequences. The objective is to find the gap placements in the multiple-sequence

alignment to give the optimum alignment score for all of the sequences at the same time.

Overall,  there  is  no  general  fast  method  for  guaranteeing  the  generation  of  optimal

multiple-sequence alignments.