Missing Junctions in Tophat! ( after providing a known junctions & gene models files)

avi · 06-01-2011, 06:37 PM

I am using Tophat for aligning unpaired RNAseq reads (36 bases each) and I had a question about junction calling in Tophat. I tried running Tophat with both -j (a known junctions file from Ensembl) and -G (known gene model annotations file from Ensembl) and also ran Cufflinks on the results.

In both of the runs, if I compare the results with the actual known transcripts from Ensembl, it seems like I am missing many known junctions.

To Illustrate, say if Ensembl has a Transcript with say 5 exons. Cufflinks annotates the same region as containing 3 transcripts (Exon 1-2 as Transcript 1, Exon 3-4 as Transcript 2 and Exon 5 as Transcript 3).

If I have understood the paper correctly, this is because even though Tophat has the coordinates of the known junctions it didn't find enough IUM reads spanning those particular junctions (under the given parameters) to allow it to call it junction? Is that correct..?

Under that assumption, I tried running Tophat with --min-anchor-length = 5 (reduced from default 8) and --min-isoform-fraction 0.1 (reduced from default 0.1). But I didn't get any improvement in finding more junctions. (The junctions.bed file has the exact same number of lines)

Does anyone have any suggestions on what else I can try to improve junction calling?

Also, given the Genomic coordinates of a splice junction, is there a way I can extract, from the Tophat output, the no of IUM (Initially Unmapped reads) that Tophat mapped to span that particular junction?

thanks,

Avinash

shurjo · 06-01-2011, 08:49 PM

Hi Avinash,

Quote:

In both of the runs, if I compare the results with the actual known transcripts from Ensembl, it seems like I am missing many known junctions.

For any one tissue type, a certain (possibly substantial) fraction of the known junctions will not be present simply due to tissue-specific expression of different isoforms. As such, I wouldn't worry about this part of your question.

Quote:

Also, given the Genomic coordinates of a splice junction, is there a way I can extract, from the Tophat output, the no of IUM (Initially Unmapped reads) that Tophat mapped to span that particular junction?

I doubt this is possible, unless you are MUCH better than hacking into the code and the tmp files than I am.

Best of luck,

Shurjo

agent99 · 08-02-2011, 04:21 PM

You could try reducing the segment-length to 1/2 of your read length. I believe that you do not get any mapped splice junctions if your segment-length is greater than 1/2 the read length.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
How to combine junctions.bed files produced by TopHat	HTS	Bioinformatics	8	05-03-2015 02:33 AM
Junctions in tophat	jay2008	Bioinformatics	0	01-26-2012 06:23 PM
tophat - no junctions	crh	Bioinformatics	2	07-25-2011 11:33 AM
junctions in tophat	agent99	RNA Sequencing	0	01-05-2011 09:51 AM
Abberant junctions by tophat	RockChalkJayhawk	Bioinformatics	3	05-05-2010 03:13 PM

06-01-2011, 06:37 PM	#1
avi Junior Member Location: Michigan Join Date: Mar 2011 Posts: 7	Missing Junctions in Tophat! ( after providing a known junctions & gene models files) I am using Tophat for aligning unpaired RNAseq reads (36 bases each) and I had a question about junction calling in Tophat. I tried running Tophat with both -j (a known junctions file from Ensembl) and -G (known gene model annotations file from Ensembl) and also ran Cufflinks on the results. In both of the runs, if I compare the results with the actual known transcripts from Ensembl, it seems like I am missing many known junctions. To Illustrate, say if Ensembl has a Transcript with say 5 exons. Cufflinks annotates the same region as containing 3 transcripts (Exon 1-2 as Transcript 1, Exon 3-4 as Transcript 2 and Exon 5 as Transcript 3). If I have understood the paper correctly, this is because even though Tophat has the coordinates of the known junctions it didn't find enough IUM reads spanning those particular junctions (under the given parameters) to allow it to call it junction? Is that correct..? Under that assumption, I tried running Tophat with --min-anchor-length = 5 (reduced from default 8) and --min-isoform-fraction 0.1 (reduced from default 0.1). But I didn't get any improvement in finding more junctions. (The junctions.bed file has the exact same number of lines) Does anyone have any suggestions on what else I can try to improve junction calling? Also, given the Genomic coordinates of a splice junction, is there a way I can extract, from the Tophat output, the no of IUM (Initially Unmapped reads) that Tophat mapped to span that particular junction? thanks, Avinash

08-02-2011, 04:21 PM	#3
agent99 Member Location: San Francisco Join Date: Jul 2010 Posts: 10	You could try reducing the segment-length to 1/2 of your read length. I believe that you do not get any mapped splice junctions if your segment-length is greater than 1/2 the read length.