Enhancing a classification job

[Amhed]

Hi everyone, hopefully someone can help me with a problem do I have...

I have Ilumina reads that were mapped against human and mouse, I run picard to separate the Aligned and unaligned reads in two SAMs files (for human and mouse), and also I wrote a perl script to obtain the first column of the SAMs file (the tags) in order to know which reads were shared betwen human and mice (I used the comm command).

So, the problem is, that I already have a txt file with the tags specific to human and mouse and I want to recover that reads from the SAM files, however it tends to be a problem of order N*M, I test something like:

grep -f tags.txt reads.sam > output.sam

However it consumes a huge amount of RAM, and the procces is killed by the sysadmin, in the other hand, I wrote a perl script that opens the SAM file and opens the tags file and compare line per line if is in the tags file but is taking so much time...

I also was planning to order the tags numerically to enhance the time, but the id is recognized as string and when I sort the tags file it gives something as:
_9985022_#CTTGTA
_9986968_#CTTGTA
_9987522_#CTTGTA
_999011_#CTTGTA
_9990461_#CTTGTA
_9991472_#CTTGTA
_9994415_#CTTGTA
_9994569_#CTTGTA
_999553_#CTTGTA
_9999187_#CTTGTA

So, if someone knows a better way to do recover the reads I really would apreciate if tell me how

Thanks in advance for your help
Amhed

[Amhed]

Solved

It seems that using a hash consumes much memory, however the process is faster enough.
Thanks