I don't have a fix for you, but instead I'm offering some refactoring advice.

1. Load what you need from the annotation file into a hash.
2. Might as well do your processing one line at a time, no need to have the whole file in memory.

Here follows some code to implement these first thoughts. I have not changed your use of the global @window, @probe, which would be my next targets for refactoring. But I think after you refactor, you may find it easier to debug.

open my $annotation_read_handle, '<', $annotation_file;
my %annotation_for;
while (my $ad = <$annotation_read_handle> ) {
    # read $an_chrom, $prol, $pror out of $ad
    my (
        $an_chrom, undef, undef, 
        $prol,     $pror, undef, 
        undef,     undef, $mess,
    ) = split(/\t/, $ad);
    # read $name out of $mess
    my (undef, undef, $name) = split(/\;/, $mess);
    
    # store for future lookups
    $annotation_for{$an_chrom} = [ $name, $prol, $pror ];
}

# loop through the main data file
OLC: 
    while (my $md = <$main_read_handle> ) {
    # remove newlines
    chomp $md; 
    # pull out chromosome #, window start, end
    my ($main_chrom, $winl, $winr) = split(/\t/, $md);
    # see if $main_chrom has been annotated
    next OLC if !exists $annotation_for{$main_chrom};
    
    my $array_ref = $annotation_for{$main_chrom};
    ( my $name, @probe ) = @$array_ref;
    # put the window start, end into array for further processing
    @window = ($winl, $winr);

    # call the range_finding sub to look for matches
    my $return = range_find();
    next OLC if !$return;
    # upon matching, print out the name of the gene along with the ori
+ginal values
    print OUTPUT "$name\t $md\n";
}
[download]

Thanks for your input. Unfortunately, this won't work as there isn't a good selection of unique identifiers to use as Keys for the Hash. So, using the code you provide, I'd end up with 23 key-value combinations, when I need 330k :). A hash of arrays would work better, in that I could have the values appended to the arrays for each chromosome, but then getting the data out would be a bit of a nightmare. I will look to cleaning up the globals though, as I was being a bit lazy there :).
EDIT: Actually, 24 combinations, as there are both x and y to consider :).

Bioinformatics

You're right. I overlooked the statement label in one of your next statements. I could not have arrived at my misreading if I had been as aware as you are that there are only two dozen chromosomes.

But what about your hash of arrays? Why would getting the data out be such a nightmare?

Populating the hash of arrays:

open my $annotation_read_handle, '<', $annotation_file;
my %annotations_for;
while (my $ad = <$annotation_read_handle> ) {
    # read $an_chrom out of $ad
    my ($an_chrom, undef) = split(/\t/, $ad);
    
    # store for future lookups
    push @$annotations_for{$an_chrom}, $ad;
}
close $annotation_read_handle;
[download]

Now read through the main data file and assign each chromosome number to my $main_chrom.

    # look up the list of annotations relevant to the current chromoso
+me
    my $annotations_ref = $annotations_for{$main_chrom};
    # loop through just these annotations
    ILC: foreach my $ad (@$annotations_ref) { # ...
    }
[download]

Of course—as other more enlightened commentators have already pointed out—the most important thing to optimize is the range_find subroutine.

I tried running the program on the data you provided but it exited instantly and printed no output in output.txt so I assume that data is not representative. I don't exactly understand what the code is doing but I don't see anything that could cause an infinite loop. I see two things that make me suspicious: 1) your range data is all numbers but you're using the character-operators lt and eq to compare them, and 2) these lines:

        # look to see if the windows don't overlap 
        if ($test lt $window[0]) {return 0;}
        # look to see if the windows don't overlap
        elsif ($test2 lt $probe[0]) {return 0;}
[download]

Update: Second thought - the second part of the code seems, if I understand it, to be designed to take two ranges of numbers and see if they overlap. To do so, it actually expands the ranges into lists and compares them one-by-one. This becomes, in the worst case, an N^2 calculation for what should be a constant operation:

sub range_overlap {
  my ($start1, $end1, $start2, $end2);
  return ($end1 >= $start2 || $start1 <= $end2) &&
         ($end2 >= $start1 || $start2 <= $end1);
}
[download]

Mmmm... The data won't match up your right. I can adjust the window values and chromosome numbers to match up so that you do get output however. I just threw up data to illustrate.
Bioinformatics

A few ways to optimise this:

1. You mentioned somewhere that hashes are not useful as there are only 24 keys. You also mentioned about using a hash of arrays, but dismissed it.

   That's premature. If you load your main file into a hash of arrays, then you only need compare each of your annotated records against 1/24th of the main records. And that reduces your overall workload and runtime to 1/24th of what it would be.

   my %main;
   my $mfile = prompt( "Main file?" );
   
   open MAIN, '<', $mfile  or die "$mfile: $!";
   
   m[^([^\t]+)\t] and push @{ $main{ $1 } }, $_
       or die "Bad record in $mfile @ $." while <MAIN>;
   
   close MAIN;
   chomp @$_ for values %main;
   [download]

2. There is no reason to load the second file into ram. You can process it one record at a time:

   my $afile = prompt( "Annotation file?" );
   open ANNOT, '<', $afile or die "$afile: $!";
   
   open OUTPUT, '>', 'output.txt' or die $!;
   [download]

   Now your main loops become:

   OUTER: while( <ANNOT> ) {
       chomp;
       my( $atag, $astart, $aend, $mess ) = split "\t";
       next unless exists $main{ $atag };
   
       for my $md ( @{ $main{ $atag } } ) {
           my( $mtag, $mstart, $mend ) = split "\t", $md;
   
           if( ?OVERLAPPED? ) {
               print OUTPUT "$atag\t$mess";
           }
       }
   }
   [download]

3. As others have pointed out, your range checking is far too complex and computationally expensive. Although there are many (7?) permutations of relative positioning of the starts and ends of the ranges, there are only 2 that do not overlap:
   1. Main end less than Annot start:

          Ms------Me
                              As------Ae
      [download]

   2. Annot end less than Main start:

                              Ms------Me
          As------Ae
      [download]

   Any other position is overlapped.

   If both your files are already in sorted order (if they are not, (learn to) use your system sort utility and sort them ascending by the tag, start, end), then when Main Start moves greater than Annot End, then you need process no further main records for this tag, so we can skip to the next Annot record. (Outer loop)

   Equally, whilst Main End is less than Annot Start, you can skip to the next Main record. (Inner loop).

   And actually, that covers both (all) conditions when we do not want to print output, so no further range testing is required.

   That makes the loops look like this:

   OUTER: while( <ANNOT> ) {
       chomp;
       my( $atag, $astart, $aend, $mess ) = split "\t";
       next unless exists $main{ $atag };
   
       for my $md ( @{ $main{ $atag } } ) {
           my( $mtag, $mstart, $mend ) = split "\t", $md;
   
           next if $mend < $astart;
           next OUTER if $mstart > $aend;
           print OUTPUT "$atag\t$mess";
       }
   }
   [download]

4. Finally, there is one further optimisation possible. As both files are sorted, when we start processing the second Annot record for a given tag, then we know it cannot be earlier than the last Annot record for this tag.

   Therefore, if we rejected and skipped the first N Main records for this tag for the last Annot record, we know we can skip at least that many for this record. This complicates the code a little, but yields a huge saving in processing which make it worth while.

The final cut of the loops code looks like this:

my( $first, $lastTag ) = ( 0, '' );
OUTER: while( <ANNOT> ) {
    chomp;

    ## ((simplified for testing)
    my( $atag, $astart, $aend, $mess ) = split "\t";
    
    ## If the tag changed, reset the start position for the Main recor
+ds
    ## and remember the new tag
    $first = 0, $lastTag = $atag if $atag ne $lastTag;
    
    ## Skip completely if this Annot tag has no corresponding Main tag
    next unless exists $main{ $atag };

    ## For each Main record with this tag,
    ## But skipping those we rejected last time
    for my $md ( @{ $main{ $atag } }[ $first .. $#{ $main{ $atag } } ]
+ ) {
        
        ## (simplified for testing)
        my( $mtag, $mstart, $mend ) = split "\t", $md;

        ## Increment the skip and goto the next Main record
        ## If the end of this Main record is less than 
        ## the start of the Annot record
        ++$first, next if $mend < $astart;
        
        ## Skip the rest of these Main record 
        ## if the start of this Main record is greater than
        ## the end of this Annot record
        next OUTER if $mstart > $aend;

        ## We have an overlap, so output the info
        print OUTPUT "$atag\t$mess";
    }
}
[download]

The result is that it now takes less that six minutes to process 300,000 record against 300,000 records, finding 113,000 overlaps:

C:\test>664760-gen > 664760.main
C:\test>wc -l 664760.main
 300000 664760.main

C:\test>664760-gen > 664760.annot
C:\test>wc -l 664760.annot
 300000 664760.annot

C:\test>664760-b
Main file?664760.main
Annotation file?664760.annot
Tue Jan 29 06:36:34 2008
Tue Jan 29 06:42:12 2008

C:\test>wc -l output.txt
 112805 output.txt
[download]

My best guess is this 3 if not 4 orders of magnitude quicker. Let's see anyone come close to that using an RDBMS.

Full test code:

The biggest suggestion I would have (after the obvious algorithmic improvement) is to pre-digest your data so that repetitive checks can be sped up or eliminated. The next thing would be to look at using a relational database (like MySQL or Oracle) and letting the power of relational calculus solve your problems.

Your right. I tried to provide checkpoints so that I wouldn't have to crawl through all the data (aka, making sure the chromosomes considered match, etc.), but even then there is a lot to go through. Heck, I even tried to write a multi threaded version of this program, but it has a few bugs still. Should it really take this long though? I don't have any benchmarks written in, so that will probably be my next step. I'll look into putting this into a database to do some of this for me, but sql is still a bit of a learning curve for me.
Bioinformatics

Do you mean the program doesn't stop even when you hit return in response to " What is the name of the ChIPOTle file\?"?

Do you mean the program never finds the end of the file it's processing?

My code "works" to some degree, in that it gives me an output

Under what circumstance does it work? For the first file only? For more than one file?

Good questions. The program works on small files, i.e. smaller annotation files of just a few lines that I used to test. They are the same format as the full files. When using the full files however, the program would not exit, even after I started it just prior to leaving on Friday, to come back to it early this morning (didn't think it would take that long :p). I did get information printed in the output file however. The program can find the end of the files, and is designed to annotate just one file at a time.
Bioinformatics

Seems to me your main block should look more like:

#!/usr/bin/perl -w
use strict;

my @main  = &open_file_main  ();
my @annot = &open_file_annot ();
open OUTPUT, '>', 'output.txt' or die "Cannot open $!";

# loop through the main data file
foreach my $md (@main) {
    chomp $md;

    my ($main_chrom, $winl, $winr) = split (/\s{2,}/, $md);

    foreach my $ad (@annot) {
        my ($an_chrom, undef, undef, $prol, $pror, undef, undef, undef
+, $mess) =
            split (/\s{2,}/, $ad);

        next unless $main_chrom eq $an_chrom;

        my (undef, undef, $name) = split (/\;/, $mess);

        next if $winr < $prol or $winl > $pror;

        print OUTPUT "$name\t $md\n";
        #last;
    }
}

exit;
[download]

Note that there is no need for a sub to open the output file! If you need to do complicated stuff in the context of opening the output file, return the file handle from the sub - do not use a 'global' variable that is set as a side effect!