Aligning only unique reads in Bowtie

[gzentner]

Hi everyone,

I am aligning some ChIP-seq data using Bowtie. I have been using the -m option to throw out any reads with > 1 reportable alignment, but I would also like to try omitting non-unique reads. Is there a command line option to throw out any reads that are identical?

Thanks!

[john_mu]

If you just have the raw reads, you can use the "uniq" command in Linux to extract the unique reads (after sorting).

http://en.wikipedia.org/wiki/Uniq

[gzentner]

Thanks John!

That sounds like it should be a useful command. I was just wondering if you could give me a little more detail.

I have the ChIP-seq data as FASTQ files which I align using bowtie. Would I use the uniq command on the FASTQ prior file to alignment to generate another FASTQ containing only unique reads?

i.e., prior to alignment, run uniq -u on the FASTQ?

Thanks!

[john_mu]

No worries, but the method I suggested is a bit of a hack... It will require you to fiddle with the data a bit.

Firstly, do you need to preserve the read-quality information? If so then it is probably best to write your own python or perl script to do it. I'm pretty sure there are existing tools to do this though... I just can't re-call off the top of my head.

-----

The method I suggested is to firstly extract the raw-reads from the FASTQ file by using
instructions here

http://www-stat.stanford.edu/~kinfai...ess.html#fastq

Then sort the reads with http://en.wikipedia.org/wiki/Sort_(Unix)

sort input_file > output_file

Finally use "uniq"

uniq -u input_file > output_file

After you do this, you can align your reads using bowtie with the "-r" option for raw reads.

[lifeng.tian]

You can try fastx_collapser from http://hannonlab.cshl.edu/fastx_toolkit/

[sridharacharya]

I have few questions regarding the best practices that are adopted, in dealing with multiple alignments from a single read and presence of identical reads in the data (from Biology stand point) :

I am curious, how important it is to deal with identical reads.
Having many identical reads in data means something wrong with the
experiment?

What could be considered as max. cutoff value for the number of identical reads in the data, so as to not consider those reads?

In the other case of a single read aligning at multiple places in a genome, what should be the cutoff value for number of multiple alignments, so as to not consider those reads?

[brojee]

No stresses, however the technique I recommended is somewhat of a hack... It will expect you to tinker with the information a bit.

Initially, do you have to save the perused quality data? In the event that in this way, at that point it is most likely best to compose your very own python or perl content to do it. I'm almost certain there are existing instruments to do this however... I just can't re-cancel the highest point of my head.