Tophat command line options

[ice]

Dear All,

I have been playing with tophat but am not certain which combination of command line options I should use. I hope people with more experiences can help me on that. The main purpose is the expression analysis and my reads are single ended with 80bp.

1. Is it better or not to provide an exon junction database? If such a database is provided, will tophat only map read against the exon junction database or it will also try the regular mapping if it fails to find a match in the database? If the answer it the latter, does that means it is always better to have such a junction database?

2. If it is better to provide such a database, where and which file to download?

3. For expression analysis should I used -g 1 to allow only unique mapping or use other values? The default is 40 and I have seen may reads reported multiple times in the sam output file. In this case, a read will be counted multiple times for different genes and this shouldn't be right.

4. The map quality scores reported in sam file by tophat have a lot of low values (e.g. >20% read have score below 20). Should I use some criteria to filter reads with bad mapping quality score? What are sensible numbers to use?

5. What other sensible parameters we should be paying attention to? For me I only used --microexon_search.

Many many thanks!

[GKM]

Originally Posted by ice

Dear All,

1. Is it better or not to provide an exon junction database? If such a database is provided, will tophat only map read against the exon junction database or it will also try the regular mapping if it fails to find a match in the database? If the answer it the latter, does that means it is always better to have such a junction database?

If you don't specifically turn off the novel junctions discovery, it should find those too. (that's the --no-novel-juncs option)

2. If it is better to provide such a database, where and which file to download?

What species are you're working with? If it's something relatively obscure, you may have to make your own junction file

3. For expression analysis should I used -g 1 to allow only unique mapping or use other values? The default is 40 and I have seen may reads reported multiple times in the sam output file. In this case, a read will be counted multiple times for different genes and this shouldn't be right.

I would keep the multireads in, you will undercount members of gene families with high sequence similarity if you look at unique reads only. Of course, there's the processed pseudogene problem too, but that's a somewhat different matter.

[sridharacharya]

Hi,

I wanted to post a new thread, but cant find "New Thread" button anywhere.

My question is :

Can Tophat handle csfasta files which are in color space format ? I think there is no command line option for doing this.
In order to work with SOLiD csfasta files, do I have to first convert it into sequence space format?

Could anyone direct me also how to post a new thread.

Thanks,

[GKM]

http://tophat.cbcb.umd.edu/manual.html

Currently, TopHat does not allow short (fewer than a few nucleotides) insertions and deletions in the alignments it reports. Support for insertions and deletions will eventually be added. TopHat also does not natively support Applied Biosystems' Colorspace format.

[ECO]

Originally Posted by sridharacharya

Hi,

I wanted to post a new thread, but cant find "New Thread" button anywhere.

My question is :

Can Tophat handle csfasta files which are in color space format ? I think there is no command line option for doing this.
In order to work with SOLiD csfasta files, do I have to first convert it into sequence space format?

Could anyone direct me also how to post a new thread.

Thanks,

See screenshot!

[ice]

Originally Posted by GKM

If you don't specifically turn off the novel junctions discovery, it should find those too. (that's the --no-novel-juncs option)

Then it is clear then that providing the junction database is clearly a win.

Originally Posted by GKM

What species are you're working with? If it's something relatively obscure, you may have to make your own junction file

My data is human RNA data. I can download the GTF file from UCSC and convert it to GFF3 format which tophat can take. Is that the right choice?

Originally Posted by GKM

I would keep the multireads in, you will undercount members of gene families with high sequence similarity if you look at unique reads only. Of course, there's the processed pseudogene problem too, but that's a somewhat different matter.

This is very interesting. Actually I am puzzled by the -g option. Suppose a read is mapped to 50 places on the genome and I specify -g 20. Will tophat (1) report only the first 20 alignments of this reads or (2) it does not report this read at all since 50 is greater than 20? It is not clear when I read the manual. If (1) is correct, then specifying a small number is better since multiple hits will over-count the expression. However if (2) is right, a lot of reads will be missed due to similarities among genes in a gene family.

[GKM]

Originally Posted by ice

Then it is clear then that providing the junction database is clearly a win.




My data is human RNA data. I can download the GTF file from UCSC and convert it to GFF3 format which tophat can take. Is that the right choice?

It depends. UCSC isn't the absolutely most comprehensive annotation, but I don't know much you care about that. You can definitely use it. GFF3 will works, as will the simple junction format that you can supply to TopHat with the -j option (chr - tab - left - tab - right - tab - strand)

This is very interesting. Actually I am puzzled by the -g option. Suppose a read is mapped to 50 places on the genome and I specify -g 20. Will tophat (1) report only the first 20 alignments of this reads or (2) it does not report this read at all since 50 is greater than 20? It is not clear when I read the manual. If (1) is correct, then specifying a small number is better since multiple hits will over-count the expression. However if (2) is right, a lot of reads will be missed due to similarities among genes in a gene family.

The -g option should be equivalent to the -m option in bowtie, i.e. if you see a read mapping to more than N locations in the genome, it is discarded.