transcriptome mapping with bowtie -- only 22% reads hit

[IrisZhu]

Following Ben's suggestion, so I mapped the two mates seperately. But still only a small percentage of reads get mapped.

===mate1=====
reads processed: 130653415
# reads with at least one reported alignment: 47875358 (36.64%)
# reads that failed to align: 82659307 (63.27%)
# reads with alignments suppressed due to -m: 118750 (0.09%)
Reported 47875358 alignments to 1 output stream(s)

====mate2=====
# reads processed: 130653415
# reads with at least one reported alignment: 35773578 (27.38%)
# reads that failed to align: 94798249 (72.56%)
# reads with alignments suppressed due to -m: 81588 (0.06%)
Reported 35773578 alignments to 1 output stream(s)

If bowtie works well, then there are two explanations: either the human refseqs are far from complete, or, the sample is contaminated with a lot of pre-mRNA/genomic DNA.
Any comments are welcome

Iris

[IrisZhu]

If she's aligning against the set of RefSeq transcripts, wouldn't she automatically get the alignments for reads spanning splice junctions?

I am supposed to. That's why I don't use tophat.

[Lee Sam]

Hi Iris,

Paired-end alignment against the transcriptome is tricky, because the distance between the mates in genome space is affected both by fragment length, and on the "shape" of the transcriptome. E.g. when a fragment spans a long intron, the intron size is in some sense "added" to the fragment length. A better way of measuring sequencing quality is to break the mates apart and align both files in an unpaired manner. And an even better way is to use TopHat or another tool that attempts spliced mapping, so that alignments that span splice junctions can be found as well.

Hope that helps,
Ben

If she's aligning against the set of RefSeq transcripts, wouldn't she automatically get the alignments for reads spanning splice junctions?

[Zigster]

A more useful way to pose this question in the future is to find one example of a read or a pair of reads that you expected would map, but did not.

[IrisZhu]

Hi Iris,

Paired-end alignment against the transcriptome is tricky, because the distance between the mates in genome space is affected both by fragment length, and on the "shape" of the transcriptome. E.g. when a fragment spans a long intron, the intron size is in some sense "added" to the fragment length. A better way of measuring sequencing quality is to break the mates apart and align both files in an unpaired manner. And an even better way is to use TopHat or another tool that attempts spliced mapping, so that alignments that span splice junctions can be found as well.

Hope that helps,
Ben

Thanks Ben.
I see. I'll map the two mates separately.

Iris

[Ben Langmead]

Hi Iris,

Paired-end alignment against the transcriptome is tricky, because the distance between the mates in genome space is affected both by fragment length, and on the "shape" of the transcriptome. E.g. when a fragment spans a long intron, the intron size is in some sense "added" to the fragment length. A better way of measuring sequencing quality is to break the mates apart and align both files in an unpaired manner. And an even better way is to use TopHat or another tool that attempts spliced mapping, so that alignments that span splice junctions can be found as well.

Hope that helps,
Ben

[IrisZhu]

To check how well the sequencing parts were done, I used bowtie to map all the reads from RNA-seq directly to human refseqs, and I got only 22% (of ) of reads have at least one hit, 78% failed to align, which I think is impossible.

my command is:
"bowtie -p 5 -m 10 -f trx -1 mate1.fa -2 mate2.fa >output"

The output:
# reads processed: 130653415
# reads with at least one reported alignment: 28765542 (22.02%)
# reads that failed to align: 101882364 (77.98%)
# reads with alignments suppressed due to -m: 5509 (0.00%)
Reported 28765542 paired-end alignments to 1 output stream(s)

Any idea why so few reads got mapped? what did I do wrong?

Thank you very much,

Iris