ChIP seq-% alignment

[SGPSling]

Hi all:

Just wondering after running the ChIP-seq data with Pipeline which by default uses ELAND for alignment, what kind of % aligment is acceptable, indicating the experiment was working fine? We got 40-60% aligment, is it reasonable? Thanks!

[Roald]

@SPGSling

What length reads are you aligning and against what reference?

Cheers
Roald

** Disclaimer: I work at CLC bio **

[SGPSling]

Our read length is 36 bp and the reference genome is human. Thanks!

[Roald]

@SPGSling

Are you allowing the algorithm to align reads with multiple matches?
If this dramatically increases the number of aligned reads it would indicate problems with overly amplified repeat regions.
Have you looked at the contigs with the aligned reads and do the data seem to make sense?

Cheers

Roald

** Disclaimer: I work at CLC bio **

[SGPSling]

We just run it with pipeline by default using ELAND to align. We are currently checking if the NM are repeats, primer dimers or other source of contamination. And we are also mapping the seq to the genome to look for peaks as well. Just want to get an idea what kind of %alignment people usually get as an indication of a good ChIP-seq experiment. Thanks a lot!

[ieuanclay]

40-60% using the default ELAND settings is a little on the low side for 36bp (paired end?). If you want to try alternate settings, bowtie (see the forum here) is very amenable to messing around with settings and also extremely fast!

Have you looked at your primer dimer occurance rate? i.e. look through your sequences for the illumina adapter sequence reverse complement (the read you would get if for an adapter/primer dimer)

[elaney_k]

I thought ChIP-seq was supposed to have a lower alignment rate than e.g. genomic DNA? We've just done some ChIP experiments and whilst we are very much still optimising the protocols our alignment was ~54% to human genome. The amount of reads containing adapter-dimer sequences (or slightly truncated ones) was <<<1%.

Anyone else willing to share their alignment percentages for human ChIP alignments?

[Chipper]

Our Illumina data are from 1G, with 30-60 % aligned reads but this number does not at all indicate how succesful the ChIP is.

[bioinfosm]

We saw chip seq on human data with about 40-50% mapped reads using eland as well..
Is there a quick way to check for adapter-dimer? Do you do it with eland itself?

[SGPSling]

We were able to sort out all the reads generated by pipeline 1.4 (default setting) for our ChIP seq experiments. This is the breakdown of what we got:

% alignment 60-70%
% repeats 20-30%
% no match 3-6%

So now we know since ELAND exclude all the reads aligned to more than 2 sites. Most of the non-aligned reads in our experiment was repearts. We did get some no-match, we are in the process to verify if it's primer dimer or other source of contaminants.

Chipper, you are absolutely right. % aligned reads does not at all indicate how succesful the ChIP is! We use it as a quality assessment of our ChIP library preparation. How good was our ChIP, we have to wait till we can validate our peaks.

[elaney_k]

We saw chip seq on human data with about 40-50% mapped reads using eland as well..
Is there a quick way to check for adapter-dimer? Do you do it with eland itself?

hi bioinfosm,

To find out the % of adapter dimers in our reads i count the number of reads containing the sequences GATCGGAAGAGCTCGTATGCCGTCTTCTGCTTG or
GATCGGAAGAGCTCGTATGCCGTCTTCT in the s_N_export.txt file using the grep command: [grep -c "GATCGGAAGAGCTCGTATGCCGTCTTCT" s_N_export.txt] and then count the total number of reads [grep -c "" s_N_export.txt].

very quick and dirty..

[olus]

I thought ChIP-seq was supposed to have a lower alignment rate than e.g. genomic DNA? We've just done some ChIP experiments and whilst we are very much still optimising the protocols our alignment was ~54% to human genome. The amount of reads containing adapter-dimer sequences (or slightly truncated ones) was <<<1%.

Anyone else willing to share their alignment percentages for human ChIP alignments?

Hi!
In a ChIP-Seq we made months ago we had 64/65% IP aligned to hg18 as from Eland Summary.htm, or 50/51% from the ShortRead report.
The latter is the percentage related to the total reads, while the first is the percentage of the reads passing the filters.
In that experiment we had a 70% of reads passing the filtering (eland default, pipeline 1.3) out of ~ 14M of clusters per lane.
We hadn't extrapolated the quota of non-aligned reads that are actually repeats, but we will...hopefully.

Cheers

[captainentropy]

In case someone else reads this, this could be of use - FastQC (http://www.bioinformatics.bbsrc.ac.uk/projects/fastqc/) analyzes your fastq dataset and will report overrepresented sequences (among other useful info). A few times I've had datasets have an adapter contamination and FastQC reported it. It even identified which adapter - in the case I'm looking at now, TruSeq Adapter 3. It was <1.1%. I'm not sure what threshold makes a sequence overrepresented though.

[olus]

In case someone else reads this, this could be of use - FastQC (http://www.bioinformatics.bbsrc.ac.uk/projects/fastqc/) analyzes your fastq dataset and will report overrepresented sequences (among other useful info). A few times I've had datasets have an adapter contamination and FastQC reported it. It even identified which adapter - in the case I'm looking at now, TruSeq Adapter 3. It was <1.1%. I'm not sure what threshold makes a sequence overrepresented though.

FastQC is a good tool. We use it routinely as a command line pipe.
With the option --casava it also recognizes the several fastq.gz splitted by the new sequencer and output one single full sample QC.


/usr/bin/fastqc --casava -t 10 ${name}_R?_???.fastq.gz

HTH