Is it OK to use paired-end 150bp to sequence ATAC-seq libraries?

[vadoue]

Hi!

I am planning to do ATAC-seq assays and I am wondering if it is fine to use paired-end 150bp to sequence ATAC-seq libraries?
Just because the sequencing platform I am working with is only doing 150bp run with HiSeq3000...
I have the impression that people are usually using PE sequencing with 50 to 75bp reads. Would the only problem be that I would sequence a lot of adapters (exept money lost, adapter trimming should be enough to solve this)? Or would I lost many "small" fragments?

Thank you for your help!
Vero

[cbp44]

Don't use large insert libraries for ATAC-seq

Hi,

This is probably too late to be of any help to you, but for anyone else doing ATAC-seq, I highly recommend doing paired-end sequencing with 50x50bp or lower.

Even with adapter trimming, I have had a very difficult time mapping a 76x76bp PE Illumina NextSeq ATAC-seq library using bowtie2. There is always a dip in mapped reads around the 76bp insert size (see attached example plot), and this has subsequently caused NucleoATAC to have trouble generating its models for nucleosome positioning analysis.

And by have trouble, I mean the program crashes when running "nucleoatac run ..." making that analysis method unusable. There was an issue created in the NucleoATAC GitHub repo that discusses this problem a little bit (https://github.com/GreenleafLab/NucleoATAC/issues/62).

Chris

[pmiguel]

You could always hard-clip your reads to 50 bases.

--
Phillip

[cbp44]

You could always hard-clip your reads to 50 bases.

Hmm... that is something I hadn't thought of actually. I am going to give that a shot to see how the insert size distribution looks after that. I'll hard clip to 50 bases, then use trim_adapters (https://github.com/ParkerLab/atactk/.../trim_adapters) to trim, but throw out reads smaller than 30bp after trimming.

I'd be interested if any other people had similar issues with an ATAC-seq library like this, and what solutions they would recommend.

I did try various adapter trimmers on the 76x76bp data and compared the mapping results. For example, I tried trim_adapters, trimmomatic and cutadapt both with and without the trimming done by bcl2fastq (so 6 different possibilities total). Using bcl2fastq trimming, and then trim_adapters seemed to work the best for me, but there wasn't much of a difference.

Most recently I tried remapping with very slow bowtie2 settings to see if that would help. For a Mus musculus sample with about 60 million paired-end reads, it took about 5 days to finish mapping (12 threads, ~2.6GHz per), but it did actually help a bit by getting me 8-10% more uniquely mapping concordant pairs, and more 70-80bp insert size reads to map.

For reference, these are the bowtie2 settings I used: "-k 4 --dovetail --no-mixed -D 20 -R 3 -N 1 -L 8 -i S,1,0.50". Mostly the same as "--very-sensitive-local", just lowering the seed size (-L), and allowing for a mis-match (-N 1).

--very-sensitive-local == -D 20 -R 3 -N 0 -L 20 -i S,1,0.50