Bowtie and Tophat in disagreement with alignment

[NM_010117]

Hi all,

I had a question about running bowtie and tophat. If I run bowtie (flags: -p 4 -S --un -v 3; paired-end) independently of tophat, I get very different results when looking at what % of the reads were mapped to my reference genome (hg18). Bowtie by itself aligns 38,474,274 of 87,783,858 reads to hg18, whereas when tophat is run (flags: -p 4 -r 20; paired-end) it maps 69,190,587 reads. Does anyone have an idea why they would differ so greatly?

The data is 76bp paired end from cancer patients done on the GAIIx.

[ZhengXia]

Tophat will map more reads than Bowtie because Tophat detects the splicing junction first which will make the reads spanning junctions mappable.

Originally Posted by NM_010117

Hi all,

I had a question about running bowtie and tophat. If I run bowtie (flags: -p 4 -S --un -v 3; paired-end) independently of tophat, I get very different results when looking at what % of the reads were mapped to my reference genome (hg18). Bowtie by itself aligns 38,474,274 of 87,783,858 reads to hg18, whereas when tophat is run (flags: -p 4 -r 20; paired-end) it maps 69,190,587 reads. Does anyone have an idea why they would differ so greatly?

The data is 76bp paired end from cancer patients done on the GAIIx.

[NM_010117]

That makes sense, but surely it shouldn't have such a large impact that it doubles the number of reads aligned to the genome?

[kmcarr]

Originally Posted by NM_010117

That makes sense, but surely it shouldn't have such a large impact that it doubles the number of reads aligned to the genome?

You observations sound perfectly in agreement with what I would expect. The average size of a exon in the human genome is 200bp. Your reads are 76nt. If the leftmost mapping position of a read is < 76bp from the end of the exon it will fail to map using bowtie. This 76bp "dead window" if you will represents 38% of the average exon length. You observed a reduction of 44%; you're in the ballbark.

[NM_010117]

Originally Posted by kmcarr

You observations sound perfectly in agreement with what I would expect. The average size of a exon in the human genome is 200bp. Your reads are 76nt. If the leftmost mapping position of a read is < 76bp from the end of the exon it will fail to map using bowtie. This 76bp "dead window" if you will represents 38% of the average exon length. You observed a reduction of 44%; you're in the ballbark.

Ah ha! Ok, that makes perfect sense. So tophat and bowtie are more or less in agreement with eachother; I'm just interpreting the results incorrectly. Well, at least it's promising to know that the mapping was done correctly at least.

Thanks for the help.