Separating ChIP from RNA-Seq data in fastq/BAM

[MikeChoud]

Hello Community!

Im finishing my PhD on epigenetics and histone PTMs in Germany and a colleague here is having a rather big problem. Replicate 2 from a transcription factor ChIP-seq (PE 150, Nova 6000) was contaminated with what we can only assume to be cDNA (Exons only, very very high number of reads). It s in the inputs and the pulldowns. The colleague sent a new batch (replicate 3) but the outcome is very very poor quality. Now replicate 2 is much better but unusable without removing the cDNA reads.

Does anyone know of a program/algorithm that can identify reads by polyT tails and remove them from the fastq/BAM file? Im rather stuck and the only solution seems to be writing a new script to do this. I find it an interesting problem from a bioinformatics point of view, I suppose it is similar in concept to separating multiplexed reads based on adaptor sequences.

Thanks for any ideas!

Cheers,
Michel

[Ben3]

MikeChoud

You can write a custom script to try and identify the polyA tails, then find the minimum TTTT and delete them. But I would say that after fragmentation, there won't be many reads coming from the cDNA with polyA tails. I don't think you can trust anything you get honestly. Wouldn't the amount of ChIP DNA be pretty low? So the amount of cDNA would probably be most of the reads anyway.

[MikeChoud]

Ben3

Yes, thats pretty much what I had in mind. But since Im a newbie in programming I was looking for a ready-made solution. That being said, my brain completely glossed over the fragmentation for RNA-seq (havent done one myself). You are absolutely right, it would be improssible to confidently remove all cDNA reads since they were fragmented. If one wants a good output, one needs a good input...

For discussions sake: The data are odd because on heatmaps you can see that the ChIP is not that bad, rep2 looks similar to rep1 but signal comparisons dont reflect that. In browser views I see that the reads are completely messed up (3000 RPKM) on highly expressed exons only but as expected on introns (0) and TFBS around promoters (150-250). So Im guessing that s the cause of bad correlation. And given the heatmap similarity and the clear differences on exons, introns and TFBS I am assuming a comperatively small cDNA contamination of otherwise not terrible TF ChIP data. Altogether it is around 8 million single mapped reads (I know it should be better).

Thank you very much for your time and thoughts, very useful indeed!

[Ben3]

MikeChoud no problem! I wish I could have helped more. Come back if you need anything else and feel free to share more about your work. It's always nice to hear what types of things people are doing in the lab and the methods they're using to solve their problems.