Can use multiple CPUs or thread to speed up the running rate for tofu_wrap.py?

lingling huang · 08-08-2016, 02:51 AM

Hi, there. I have 7 smrtcell data. It's really slow to run tofu_wrap.py, I want to know to weather I can use multiple CPUs or thread to speed up the running rate. Thank you for any tips!

Magdoll · 08-08-2016, 10:03 AM

Hi,

Short answer is yes. There are multiple ways to hack tofu_wrap.py -- will require additional monitoring of the parallel jobs. But I first need to ask what parameters did you use to call tofu_wrap.py? What are the size bins (aka what are the subdirectories in clusterOut/)? Do you have an SGE cluster? Do you have multiple nodes from which you can run parallel pbtranscript.py cluster jobs?

Also, please consider joining the Iso-Seq google group: https://groups.google.com/forum/#!forum/smrt_isoseq

Since tofu_wrap.py is on the cutting edge (it's not officially supported in SMRTAnalysis 2.x but is on the agenda for SMRTAnalysis 3.x), the google group is better suited

--Liz

lingling huang · 08-08-2016, 08:49 PM

Quote:

Originally Posted by Magdoll

Hi,

Short answer is yes. There are multiple ways to hack tofu_wrap.py -- will require additional monitoring of the parallel jobs. But I first need to ask what parameters did you use to call tofu_wrap.py? What are the size bins (aka what are the subdirectories in clusterOut/)? Do you have an SGE cluster? Do you have multiple nodes from which you can run parallel pbtranscript.py cluster jobs?

Also, please consider joining the Iso-Seq google group: https://groups.google.com/forum/#!forum/smrt_isoseq

Since tofu_wrap.py is on the cutting edge (it's not officially supported in SMRTAnalysis 2.x but is on the agenda for SMRTAnalysis 3.x), the google group is better suited

--Liz

Code:

tofu_wrap.py --nfl_fa isoseq_nfl.fasta --ccs_fofn reads_of_insert.fofn --bas_fofn input.fofn -d clusterOut --quiver --bin_manual "(0,2,4,6,8,9,10,11,12,13,15,17,19,20,23)" --gmap_db /zs32/data-analysis/liucy_group/llhuang/Reflib/gmapdb --gmap_name gmapdb_h19 --output_seqid_prefix human isoseq_flnc.fasta final.consensus.fa

The lab's sever is high-powered single-node computer and
has no an SGE cluster. Can I use multiple CPUs to run command?

Magdoll · 08-09-2016, 03:11 PM

Without SGE, it will be slow anyway.

But if you think that single node can handle it, one way is to run multiple instances of `pbtranscript.py cluster` on the different bins.

ex: tofu_wrap.py always creates bins 0to1kb_part0, 1to2kb_part0, etc

You can terminate tofu_wrap and keep the bins as they are. Then separately in each bin call a separate instance of cluster:

pbtranscript.py cluster isoseq_flnc.fasta final.consensus.fa \
--nfl_fa isoseq_nfl.fasta -d cluster --ccs_fofn reads_of_insert.fofn \
--bas_fofn input.fofn --quiver --use_sge \
--max_sge_jobs 40 --unique_id 300 --blasr_nproc 24 --quiver_nproc 8

(for you, you would remove the --use_sge and --max_sge_jobs option)

(see cluster tutorial here: https://github.com/PacificBioscience...-and-Quiver%29)

My guess is this would make it a bit faster but still relatively slow since everything is running in serial instead of parallel, but may be better than nothing...

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
hisat2 multiple thread usage	ronaldrcutler	RNA Sequencing	6	06-16-2016 10:12 AM
High multiple mapping rate Drosophila pair-end 50 bp RNAseq	TTYE	Illumina/Solexa	7	06-03-2015 06:22 AM
Thoughts on new workstation/server using Haswell CPUs	atcghelix	Bioinformatics	18	09-04-2014 12:53 PM
Should one instance of mpileup use a full thread? Computer running slowly.	Heisman	Bioinformatics	6	07-19-2013 05:12 AM

08-08-2016, 02:51 AM	#1
lingling huang Member Location: changsha Join Date: Mar 2016 Posts: 46	Can use multiple CPUs or thread to speed up the running rate for tofu_wrap.py? Hi, there. I have 7 smrtcell data. It's really slow to run tofu_wrap.py, I want to know to weather I can use multiple CPUs or thread to speed up the running rate. Thank you for any tips!

08-08-2016, 10:03 AM	#2
Magdoll Member Location: Bay Area Join Date: Aug 2011 Posts: 30	Hi, Short answer is yes. There are multiple ways to hack tofu_wrap.py -- will require additional monitoring of the parallel jobs. But I first need to ask what parameters did you use to call tofu_wrap.py? What are the size bins (aka what are the subdirectories in clusterOut/)? Do you have an SGE cluster? Do you have multiple nodes from which you can run parallel pbtranscript.py cluster jobs? Also, please consider joining the Iso-Seq google group: https://groups.google.com/forum/#!forum/smrt_isoseq Since tofu_wrap.py is on the cutting edge (it's not officially supported in SMRTAnalysis 2.x but is on the agenda for SMRTAnalysis 3.x), the google group is better suited --Liz

08-09-2016, 03:11 PM	#4
Magdoll Member Location: Bay Area Join Date: Aug 2011 Posts: 30	Without SGE, it will be slow anyway. But if you think that single node can handle it, one way is to run multiple instances of `pbtranscript.py cluster` on the different bins. ex: tofu_wrap.py always creates bins 0to1kb_part0, 1to2kb_part0, etc You can terminate tofu_wrap and keep the bins as they are. Then separately in each bin call a separate instance of cluster: pbtranscript.py cluster isoseq_flnc.fasta final.consensus.fa \ --nfl_fa isoseq_nfl.fasta -d cluster --ccs_fofn reads_of_insert.fofn \ --bas_fofn input.fofn --quiver --use_sge \ --max_sge_jobs 40 --unique_id 300 --blasr_nproc 24 --quiver_nproc 8 (for you, you would remove the --use_sge and --max_sge_jobs option) (see cluster tutorial here: https://github.com/PacificBioscience...-and-Quiver%29) My guess is this would make it a bit faster but still relatively slow since everything is running in serial instead of parallel, but may be better than nothing...