Re^5: counting the number of 16384 pattern matches in a large DNA sequence

[BrowserUk]

Unless I'm missing something, ... (as long as it's not necessary to match overlapping matches):

You hit the nail on the head. You'll only match 5,015,229 times when the OPs code matches 35,106,546 times.

However, with a modification to your regex, you can avoid that problem and find overlapping matches:

++$index{ $1 } while $$rSeq =~ m[(?=([ACGT]{7}))]g;

But it is still much slower than avoiding the regex engine completely:

[ 6:55:26.00] C:\test\humanGenome>..\976237 chr21.fa
16384
Using custom indexing found 35106546 matches; took 31.611258 seconds
Using custom index2 found 35106546 matches; took 27.504099 seconds
Using custom index3 found 35106546 matches; took 27.571143 seconds
Using quantified charclass lookahead found 35106546 matches; took 49.9
+54810 seconds

But ++ for thinking outside the box. (I can't believe I actually used that phrase :)

[salva]

an untested variation:

while(/([ACGT]{7,})/g) {
  for my $ix (0..lenght($1) - 7) {
    ++$index{substr($1, $ix, 7)}
  }
}

This regular expression should process every character on the string just once and so be an order of magnitude faster than yours which tries to match the look-ahead pattern at every char.

But that is just guessing... could you benchmark it?

[aaron_baugher]

Funny, I thought I was getting back inside the box by using a simple regex. :-)

I've never been totally comfortable with the zero-width assertions, because logically it seems to me that that should keep matching in the same spot. If zero-width really means zero-width, the next match should start at the same point and match the same string again, shouldn't it? I guess the semantics of it just bug me. I feel like I should have to use something like this to make sure it advances a character:

m[([ACGT](?=[ACGT]{6}))];

Anyway, I'm impressed that the regex does as well as it does, considering how simple it was to implement.

[BrowserUk]

If zero-width really means zero-width, the next match should start at the same point and match the same string again, shouldn't it?

Having the regex engine always move on by at least 1 character is a way to ensure that it always makes progress. Thus it prevents many cases of endless looping that would result in pathological behaviors.

Once you know that, the semantics are very useful.

Conversely this m[([ACGT](?=[ACGT]{6}))]; only captures a single character each time. Unless the capturing parens are inside the lookahead construct, anything they match is not captured.

Again, once you know the (slightly counter-intuitive) semantics, this proves to be quite useful.