need some help with Tophat

[IrisZhu]

Hello All,

I am new to this site and also new to the field of next generation seq.
I am trying to use Tophat (version 1.0.14) to do RNA-seq mapping. And I found it pretty slow: It took ~15 hours to finish the mapping of ~17 million pairs of reads(75bp) against a chr20 index. My reads are in fasta format and the command I used is:
"tophat -r 250 chr20 **.lane1_1.fa **.lane1_2.fa"

I think I installed Tophat and Bowtie properly since they passed the test. But I must have done something silly. I can't imagine how long it's gonna take if I map 400 million reads against the whole human genome.

Could any one tell what I could do wrong?

Thank you so much,

Iris

[john_mu]

That is not slow. It sounds about the right time.

Also, yes 400million reads will need a powerful multi-core computer with a lot of memory.

For example, it took me about 17 hours to map 130 million (100bp) reads with SpliceMap on 10 cores.

The current version of SpliceMap is also about the same speed as TopHat. However, the next version will be twice as fast.

Edit: use the -p option in TopHat if you have a multi-core machine.

[IrisZhu]

John, thanks for your reply.

This 15 hours for mapping 17 million pairs against a single (short) chromosome ( chr20 ) is reasonable? I am asking just to make sure you was not thinking of the whole genome :-)

[john_mu]

John, thanks for your reply.

This 15 hours for mapping 17 million pairs against a single (short) chromosome ( chr20 ) is reasonable? I am asking just to make sure you was not thinking of the whole genome :-)

oh... a single chromosome??? That is uhhh... a bit slow.

How did you build the index?

EDIT: also how much free memory do you have? Maybe either bowtie or tophat is thrashing.

[IrisZhu]

Now I am mapping the same thing against the whole genome index, seems it's not a lot more slower than just against a chr20 index ......

Another thing, if I add option "-p" it failed immediately:
"
[zhuz2@cbbdev1 mapping]$ tophat -p -r 100 hg19 mate1.fa mate2.fa
Traceback (most recent call last):
File "/usr/local/tophat/1.0.14/bin/tophat", line 1854, in <module>
sys.exit(main())
File "/usr/local/tophat/1.0.14/bin/tophat", line 1746, in main
args = params.parse_options(argv)
File "/usr/local/tophat/1.0.14/bin/tophat", line 474, in parse_options
self.system_params.parse_options(opts)
File "/usr/local/tophat/1.0.14/bin/tophat", line 171, in parse_options
self.bowtie_threads = int(value)
ValueError: invalid literal for int() with base 10: '-r'
"
The command "tophat -r 100 hg19 mate1.fa mate2.fa" works well.
Do you know why? Is it the problem of my machine or command?

Thank you so much for your help,

Iris

[john_mu]

you need to use -p followed by the number of processes.

Now I am mapping the same thing against the whole genome index, seems it's not a lot more slower than just against a chr20 index ......

Another thing, if I add option "-p" it failed immediately:
"
[zhuz2@cbbdev1 mapping]$ tophat -p -r 100 hg19 mate1.fa mate2.fa
Traceback (most recent call last):
File "/usr/local/tophat/1.0.14/bin/tophat", line 1854, in <module>
sys.exit(main())
File "/usr/local/tophat/1.0.14/bin/tophat", line 1746, in main
args = params.parse_options(argv)
File "/usr/local/tophat/1.0.14/bin/tophat", line 474, in parse_options
self.system_params.parse_options(opts)
File "/usr/local/tophat/1.0.14/bin/tophat", line 171, in parse_options
self.bowtie_threads = int(value)
ValueError: invalid literal for int() with base 10: '-r'
"
The command "tophat -r 100 hg19 mate1.fa mate2.fa" works well.
Do you know why? Is it the problem of my machine or command?

Thank you so much for your help,

Iris

[IrisZhu]

Yes I just realized this a few minutes ago :-) so silly of me ....
Thanks again.

[poisson200]

Hi,
Just an observation; is there a reason to map only to chromosome 20? It is possible some reads may map to chromosome 20 but better map to a different chromosome (less mismatches). So some reads may be falsely mapped to Chromosome 20 if the others are not present in the indexes. Also you would not know if reads mapped to more than one chromosome with the same stringency. Possibly something to bare in mind.

[IrisZhu]

Hi,
Just an observation; is there a reason to map only to chromosome 20? It is possible some reads may map to chromosome 20 but better map to a different chromosome (less mismatches). So some reads may be falsely mapped to Chromosome 20 if the others are not present in the indexes. Also you would not know if reads mapped to more than one chromosome with the same stringency. Possibly something to bare in mind.

Thanks for your comment. Of course it doesn't make sense to map to one chromosome.
That's just for a test to see if the software is working.

[poisson200]

In that case, it makes perfect sense.