[n00b question] best way to make whole genome windowed coverage correlation?

[zaratieg]

Hi everyone

I have three samples that i would like to compare for differences in coverage. Right now the data is in .bed files. I'd like to run some statistics on them but i'm afraid I don't know how to do that. My questions are the following.

1. What is the best normalization scheme for coverage counts, to account for library size/total read count?
1. Is there software to do the correlation quickly?
3. I'm also trying to learn R. I can import the .bed files to GRanges, but after that I'm stumped. Anyone know a good BioConductor package for this kind of thing?

Thank you in advance

[dpryan]

The normalization strategy will depend a bit on the underlying nature of the experiment. If you expect most of the genomic chunks that you're looking at to be the same between samples then the procedures from DESeq or edgeR would work OK. R can generally calculate correlations pretty quickly. Did you just convert the mapped reads to BED format and then import that or are these regions with associated counts (i.e., modified BED files)?

[zaratieg]

Thanks for your answer Devon. The data are viral insertion sites, so most of the genome is empty, with around 3000 clusters with signal ranging from 1 to 300 counts. The mapped read tells me the insertion point at its 5' end, so I made bedgraph files from the bam files with bedtools genomecov -5, and bed files with bamtobed and piping it through awk to shorten the alignments to the 5' end. The library sizes were relatively similar but not enough to ignore normalization (2.0e6, 1.92e6 and 1.83e6).
For the normalization, aren't DEseq and edgeR used for RNAseq? as you can see this is probably a very different problem.