[twu]

Meaning of -n and -Q

The -n flag limits the number of results reported, not the number of alignment results. Suppose you specify "-n 1", and GSNAP finds no alignments to the genome. Then the result would contain 0 hits, obviously. Likewise, if GSNAP finds 1 hit in the genome, the result would contain that one hit. But if GSNAP found multiple hits, then the "-n 1" flag would constrain the results to a single one.

To find uniquely matching hits, you would need to add the "-Q" or "--quiet-if-excessive" flag. With "-n 1" and that flag, if GSNAP found multiple hits, then it would pretend that it really found no (unique) hits to the genome, and report no hits.

I hope that makes sense. If you have other questions, you can also join the gsnap-users mailing list at EBI.

Tom

[ppoudel]

Thanks Parthav and Thomas!!!

I have ~5 GB (each) of paired end reads from RNA seq experiment and I would like use GSNAP in HPC farm, however, the farm allows 12 hours maximum for a job. I tried to use 1 nodes and 16 process , time=12 hours and memory 64000. But the job crashed. Is there any way I can run this in HPC within 12 hours.

[twu]

Running GSNAP on a farm

For HPC or Linux farms, it is probably best if you run GSNAP on several nodes for a given input file. You can do this easily with the -q or --part flag, which breaks up the input into parts. For example, if you want to spread the GSNAP computation over 50 nodes, you can run GSNAP like this:

gsnap -q 00/50 <fastq file> > output.00
gsnap -q 01/50 <fastq file> > output.01
gsnap -q 02/50 <fastq file> > output.02
...
gsnap -q 49/50 <fastq file> > output.49

The meaning of "-q 02/50" is that in every set of 50 input reads, compute on the second one. If you submit each of the above jobs to a different node on your cluster, you should theoretically see a 50x speedup. However, your output will be spread among 50 different files.

Note that I am working on making GSNAP a bit faster, but adding an initial alignment step that computes the easy alignments very quickly, but falls back upon the existing algorithm to harder alignments.

Regards,

Tom

[ParthavJailwala]

Tom,

Thanks for your input on how to break up the input fastq file into parts for using multiple HPC nodes, using the "-q" flag.
In case of a gsnap mapping run involving paired-end fastq reads, I am wondering how the '-q' works. How would one specify picking the read-pairs from the R1 and R2 files?

Thanks
Parthav

For HPC or Linux farms, it is probably best if you run GSNAP on several nodes for a given input file. You can do this easily with the -q or --part flag, which breaks up the input into parts. For example, if you want to spread the GSNAP computation over 50 nodes, you can run GSNAP like this:

gsnap -q 00/50 <fastq file> > output.00
gsnap -q 01/50 <fastq file> > output.01
gsnap -q 02/50 <fastq file> > output.02
...
gsnap -q 49/50 <fastq file> > output.49

The meaning of "-q 02/50" is that in every set of 50 input reads, compute on the second one. If you submit each of the above jobs to a different node on your cluster, you should theoretically see a 50x speedup. However, your output will be spread among 50 different files.

Note that I am working on making GSNAP a bit faster, but adding an initial alignment step that computes the easy alignments very quickly, but falls back upon the existing algorithm to harder alignments.

Regards,

Tom

[twu]

-q flag and paired-end reads

If you have paired-end reads (by providing two files to GSNAP), then the -q flag takes the correct pairs from each of the files. That's the only thing that makes sense.

For example, -q 2/50 takes the second read out of each set of 50, from each of the two files.

Tom

[ppoudel]

Thanks Tom.

[amolkolte]

I have been running GSNAP as a part of trinity pipeline.

To begin with, I have paired ends reads of 7GB of size for each read file.
The pipeline has executed the following code,

gsnap -d target.gmap -D . -A sam -N 1 -w 10000 -n 20 -t 45 /home/amol/trimmed_datasets/AP_treated/R33_APT_s_3_1_trimmed.fastq /home/amol/trimmed_datasets/AP_treated/R33_APT_s_3_2_trimmed.fastq

More that 30 hours have been passed and the script is showing a message "Starting alignment". I can see (using top) that, gsnap is taking resources but at the same time its showing the process status as 'sleeping'

Can anyone please explain.

Thanks

[amolkolte]

After running for 6 days, gsnap finally generated an 34 GB sam file.

Alternatively, I noticed at later point that you can monitor the output file size under your 'gsnap_out' directory and estimate where the alignment have reached.

du -h gsnap_out/gsnap.sam

File size would increase as the alignment proceeds and you can make sure that the program is running.