Tophat: confusing results with exon junctions

[goudurix]

Dear all,
I got extremely confusing results related to exon-exon junction reports with tophat.

I'm interested in finding exon-exon juntions in new genes. When using tophat with GTF annotation it map read to known junctions but fail to find any junctions in previously uncharacterized genes.

My dataset (illumina single-end PolyA RNA-seq experiment) contains ~85 millions quality checked reads whose the mode of length distribution is 36mer (file merge.fastq). In order to test tophat performance I selected a subset (5 millions reads) of my read dataset (file merge.5M.fastq).

Using tophat without GTF, I got 19992 junctions with the subset. In contrast, using the same parameters, I got less junctions with the full dataset (18547) and lots of correct junctions that were found with the subset were not found with the full dataset. What is extremely intriging is that the two junction.bed files only contain 50 junctions in common.
I am very confused with this results. Although I would suspect a bug somewhere there is perhaps a parameter that should be tuned (I have already tested several tuning of --segment-length).

Any Idea is welcome.

Thanks for help
Cheers


# commands copied from run.log:
# Full dataset

/usr/local/bin/tophat -p 5 --segment-mismatches 1 --segment-length 28 --min-anchor-length 7 --library-type fr-secondstrand --keep-tmp -o tophat_results_full /genomes/Mus_musculus/UCSC/mm9/Sequence/Bowtie2Index/mm9 merge.fastq

# commands copied from run.log:
# subset (5M)

/usr/local/bin/tophat -p 5 --segment-mismatches 1 --segment-length 28 --min-anchor-length 7 --library-type fr-secondstrand --keep-tmp -o tophat_results_segLen_28_minAnchorLength_7_segMisMatch_1 /genomes/Mus_musculus/UCSC/mm9/Sequence/Bowtie2Index/mm9 merge.5M.fastq

# Software version
/usr/local/bin/tophat -v
TopHat v2.0.4

/usr/local/bin/prep_reads
prep_reads v2.0.4 (3480M)

/usr/local/bin/segment_juncs
segment_juncs v2.0.4 (3480M)

/usr/local/bin/juncs_db
juncs_db v2.0.4 (3480M)

/usr/local/bin/bowtie2-build
Bowtie 2 version 2.0.0-beta7

/usr/local/bin/long_spanning_reads
long_spanning_reads v2.0.4 (3480M)

[goudurix]

Still no reply ?
I havent' found any fix...
(RNA-seq is strand-specific)

[Jianguang]

Hi Goudurix,
Hope you have already fixed the problem.
I am new to RNA-seq analysis and come to ask you a question.
You mentioned that you ran Tophat with only a subset of your dataset. How did you do that. I am using the web-based Galaxy system, I appreciate it very much if you have time to let me know how to make a small dataset with a subset of parent dataset in Galaxy.
Thanks,
Jianguang