DNAse-seq/Bowtie > Many non-unique alignments???

[dedee]

Hi all,

I have analysed ChIPseq data before, but have no experience DNAse-seq, so any input is greatly appreciated

I have DNA-se seq data from a Solexa GA (from a collaborator). For reasons that are beyond me I got the raw data in bam format. But well.
Trimmed the 50bp reads to 20bp, fed them into bowtie (-m1) and got sth like 95% of reads with more than one alignment.
Is this normal?

Thanks!
dedee

[kmcarr]

dedee,

What organism are you working in? What other options did you use for bowtie? Why did you trim the reads to just 20bp?

[colindaven]

Sounds like the read trimming is the problem. I seem to remember with 36-40bp Illumina reads I get many non-specific hits in a human transcriptome alignment (>80% I think!).

Perhaps an alternative, not so harsh approach is to align (preferably long sequences, perhaps using bwa) then filter by Mapping quality afterwards.

[dedee]

Sorry, thanks for answering anyway.

20bp: the quality scores for everything above 20bp look sh**, so even my collaborator stated that everything more than 20bp is not informative - and suggested to trim down the reads. Which I am trying.

Organism: mouse, mm9
I've tried aligning with -v0 (no mismatches allowed) and this gives some more sensible results:

bowtie -m1 -v0 --sam mm9 filename.sam
[samopen] SAM header is present: 22 sequences.
# reads processed: 55341291
# reads with at least one reported alignment: 42077875 (76.03%)
# reads that failed to align: 3731393 (6.74%)
# reads with alignments suppressed due to -m: 9532023 (17.22%)

Sorry for the hassle - I was asking for help as figuring that out myself can (worst case scenario) take forever. I'm posting the result/solution in case anybody else (with little experience) is running into the same issue.
Thanks!
D

[dedee]

Sounds like the read trimming is the problem. I seem to remember with 36-40bp Illumina reads I get many non-specific hits in a human transcriptome alignment (>80% I think!).

Perhaps an alternative, not so harsh approach is to align (preferably long sequences, perhaps using bwa) then filter by Mapping quality afterwards.

Thanks!
I'm trying this approach as well, using the entire 50bp reads. It just takes a while...
In the meantime, allowing 0 mismatches with bowtie and the 20bp trimmed reads gave me sth that looks about right to me. I'll report further.

[kmcarr]

Sorry, thanks for answering anyway.
...
Sorry for the hassle - I was asking for help as figuring that out myself can (worst case scenario) take forever. I'm posting the result/solution in case anybody else (with little experience) is running into the same issue.
Thanks!
D

No hassle, just wanted to get a better picture of the experiment to advise.

But I agree with colindaven and your subsequent results support that. Aligning 20mers with errors permitted to a complex eukaryote, with repetitive DNA doesn't provide enough uniqueness.

And as an unsolicited piece of advice, if your data is s*^t after 20 bp is it worth using at all? Generating sequence these days is cheap. Time spent analyzing it is expensive.