ChIP-seq library pre-processing

[BatSeqs]

Hello everyone,

I am really hoping to find some ChIP-seq seasoned analysers around

I have several questions but the first one is about the pre-processing of reads, before mapping to a reference genome.

I have publicly available data from modENCODE, against all the "standards" in the ENCODE and derived projects, there is only one replicate of the experiment, and the input library seems to be shared between all the ChIP experiments performed by that lab in that Drosophila strain/developmental time. Anyway, I thought it would be a nice (no merging the replicates issues) data set to practice the analysis of ChIP-seq Peaks.

My main concern is related to the artefacts one may introduce when filtering or processing for quality. I will explain my particular situation with the following points. After you perform a Quality Check (i.e. typically with FastQC) and discover that:

a) the quality of your reads decreases at the end with each cycle

In my case the Input library has an awful per base sequence quality but the ChIP library doesn't



This means that I should trim my reads in the Input library but not the reads in the ChIP library (right?). What I would do is trim the reads to maintain a Phred Score greater than some value for each base call, and then, filter out reads that are shorter than say, 18 bp.

b) Overrepresented sequences are an issue in the library
Which by the way, brings me to a question: wth happens when the overrepresented sequences have no hit (meaning, they aren't adaptors, right?), are these the result of a very exaggerated amplification?
Anyway, one is supposed to filter those out too, right?

As I see it, modifying the libraries because of the previously mentioned reasons leads to:
*Modifying reads
*Filtering out reads

In the first case, as long as they are mappable is OK, but in the second case, how can we possibly remove reads when the ChIP-seq methodology is based on the accountability of sequences (reads). I can think for example that if one eliminates reads from the Input library, when the time comes to compare the enrichment over the input, artificial enrichments will be found because the reads that are supposed to be there and indicate "this is NOT a peak" are not there anymore.... or the other way around, one could think there is definitely a depletion of protein binding in a place where reads from the ChIP library are filtered out, but not from the input.

HOW CAN WE POSSIBLY TREAT EACH ChIP-seq LIBRARY INDEPENDENTLY?

Thanks!

[Brian Bushnell]

Illumina reads always decrease in quality after each cycle. You should trim the adapters prior to mapping, using a trimming tool such as BBDuk. It's also possible that your unwanted peaks are caused by known artifacts, in which case it would be helpful for you to describe their sequence in this thread.

[idonaldson]

I recommend using trimmomatic filtering for adaptor sequences and trim by quality. HiSeq and MiSeq use TruSeq3 files, whereas the older sequences use TruSeq2 files; check the manual. The adapter files are also dependent on whether the sequencing was single-ended or paired. I would use the sliding window trimmer with a Q20 cut off; the window size params seem to work fine for me. I would also remove reads less than 25bp. You can always run fastqc again before and after mapping to see the quality of unmapped and mapped reads. Apart from that the success of the peak calling will depend on the coverage (number of reads). also try MACS2.1.0 for peak calling.

I hope this helps.