similar string matching

Anonymous Monk has asked for the wisdom of the Perl Monks concerning the following question:

Hi monks,

I have at my work a little tricky task. It is a string matching problem. I want to compare two amino acid sequences (here in Letter code, single char represents one amino acid) to find at which positions one sequence lies in the other! example?

MAAGAAAAFAAAATTTTTTTTFTTTTTTTTTTTTAAAAEAAAARAAAAAA # 1. sequence
             TTTTTTTTFTTTTTTTTTTTT                 # 2. sequence
result is: 2. lies at position 14 to 34 in 1.
[download]

simple? new examples

SUBSTITUTION
AAAAEAAAARGAAATTTTFTTTTTTTTTTTTTTTTAAAAAAAAILVAAAAAAAA # 1. sequence
             TTTTFTTTATTTTTTDTTTTT                 # 2. sequence
DELETION
AAAAAAAAAAAAATTGTTTTTTTXXXXXTTTTTTTTTTMAAAAAAAAAAAAAAAA # 1. sequence
             TTGTTTTTTTTTTTTTTTTTM                 # 2. sequence

REVERSE
               TTTTTTTTTTTTTTTTTTTT  # 1. sequence
  AAAAAAAAAAAATTTTTTTTTTTTTTTTTTTTTAAAAAAAAAAAAAAAA # 2. sequence

PERFECT MATCHING ONLY AT BEGIN AND END OF 2. SEQUENCE
AAAAAAAAAAATTTTTTTTGGGGGGGGGGGGGGGGGGGGGTTTTTTTTTAAAAAAA # 1.sequence
           TTTTTTTTGGGNNGGGEEGGGEGGGGGGTTTTTTTTT    # 2. Sequence
[download]

I tried with the regexp and the module String::Approx and aslice with the option 'minimal_distance', but I don't like the return values for this module.

Any hints how to do "the best way"?

Murcia

edit (broquaint): dropped <pre> tags aand added formatting

Comment on similar string matching Select or Download Code

Replies are listed 'Best First'.
Re: similar string matching by hv (Prior) on Jul 05, 2004 at 12:12 UTC
I don't know about "best way" - I suspect String::Approx is about as good as we have right now. Here's an approach that may give some food for thought though: `# assuming $left sequence is the longer one my $len = length($right); my @matches = map { (substr($left, $_, $len) ^ $right) =~ tr/\0/\0/ } 0 .. length($left) - $len;` [download] This takes advantage of the fact that the exclusive-or of two characters is zero iff the two characters are the same, to give a list of the number of matching characters when you line `$right` up against each position of `$left`. You may also want to take a look at the recent thread non-exact regexp matches, particularly the references to Text::Levenshtein and List::Compare - the author there appears to be trying to solve a harder problem, so you may find what you want falling out trivially from one of the proposed solutions there. Hugo	[reply] [d/l] [select]
Re: similar string matching by BrowserUk (Patriarch) on Jul 05, 2004 at 13:41 UTC
Needs lots more testcases and no promises about speed, but this seems to work for the cases you outlined. Results: (underscores show the offset; lowercase left partial match; uppercase right partial match. P:\test>371799 Searching: MAAGAAAAFAAAATTTTTTTTFTTTTTTTTTTTTAAAAEAAAARAAAAAA for : TTTTTTTTFTTTTTTTTTTTT MAAGAAAAFAAAATTTTTTTTFTTTTTTTTTTTTAAAAEAAAARAAAAAA _____________TTTTTTTTFTTTTTTTTTTTT Searching: AAAAEAAAARGAAATTTTFTTTTTTTTTTTTTTTTAAAAAAAAILVAAAAAAAA for : TTTTFTTTATTTTTTDTTTTT AAAAEAAAARGAAATTTTFTTTTTTTTTTTTTTTTAAAAAAAAILVAAAAAAAA ______________ttttfttttttttttttttTT Searching: AAAAAAAAAAAAATTGTTTTTTTXXXXXTTTTTTTTTTMAAAAAAAAAAAAAAAA for : TTGTTTTTTTTTTTTTTTTTM AAAAAAAAAAAAATTGTTTTTTTXXXXXTTTTTTTTTTMAAAAAAAAAAAAAAAA _____________ttgttttttt-----TTTTTTTTTTM Searching: TTTTTTTTTTTTTTTTTTTT for : AAAAAAAAAAAATTTTTTTTTTTTTTTTTTTTTAAAAAAAAAAAAAAAA AAAAAAAAAAAATTTTTTTTTTTTTTTTTTTTTAAAAAAAAAAAAAAAA ____________TTTTTTTTTTTTTTTTTTTT Searching: AAAAAAAAAAATTTTTTTTGGGGGGGGGGGGGGGGGGGGGTTTTTTTTTAAAAAAA for : TTTTTTTTGGGNNGGGEEGGGEGGGGGGTTTTTTTTT AAAAAAAAAAATTTTTTTTGGGGGGGGGGGGGGGGGGGGGTTTTTTTTTAAAAAAA ___________ttttttttgggggggggggggggGGGGGGTTTTTTTTT [download] Code Read more... (3 kB)	[reply] [d/l] [select]
Re: similar string matching by dakkar (Hermit) on Jul 05, 2004 at 16:08 UTC
The algorithms are quite well established, but they're not obvious. If you try to roll your own, you'll quite probably end up with very slow matching. Look at http://www.bioperl.org/ and you'll find all the information you need. The short version: use BLAST or derived methods. -- dakkar - Mobilis in mobile Most of my code is tested... Perl is strongly typed, it just has very few types (Dan)	[reply]
Re^2: similar string matching by Anonymous Monk on Jul 05, 2004 at 19:22 UTC
Nice hint but what if I want to find single amino acid sequences? like cystine bridges? Murcia	[reply]
Re^3: similar string matching by dakkar (Hermit) on Jul 05, 2004 at 19:41 UTC
Ehm.. I'm not a biologist, I only know someone... so you almost lost me there. The BLAST and related algorithms do exactly what you asked for: they find all the possible matches between two sequences, ranking them by 'edit distance', i.e. the number of operations needed to obtain a perfect match. This is really all I know, for details ask an expert `;-)` -- dakkar - Mobilis in mobile Most of my code is tested... Perl is strongly typed, it just has very few types (Dan)	[reply]
Re: similar string matching by educated_foo (Vicar) on Jul 05, 2004 at 14:05 UTC
What you want is the unanchored version of String::Approx. In other words, what it does, except with zero cost for insertions at the start or end of the small sequence. The C code of S::A is a bit hairy, so modifying that might not be the best way. A google search for "local alignments" should turn up lots of good references -- bioinformatics is well-documented online if you know the right keywords.	[reply]
Re: similar string matching by chiburashka (Initiate) on Jul 05, 2004 at 07:55 UTC
from what i understand : `$x = "MAAGAAAAFAAAATTTTTTTTFTTTTTTTTTTTTAAAAEAAAARAAAAAA"; $y = "TTTTTTTTFTTTTTTTTTTTT"; if ($x =~ /$y/o) {print "the second amino acid is in the first";} if ($y =~ /$x/o) {print "the first amino acid is in the second";}` [download]	[reply] [d/l]
Re^2: similar string matching by Murcia (Monk) on Jul 05, 2004 at 09:15 UTC
That is the simplest case!	[reply]


laziness, impatience, and hubris
	PerlMonks