search

RegulatoryGenomicsGroup / chicdiff

A differential caller for capture Hi-C data
4 stars 3 forks source link

Semantic problem: interaction-level normalization factor computation #7

Open armenabnousi opened 4 years ago

armenabnousi commented 4 years ago

Hi,

Thank you for this awesome package! This is not a problem with the package itself, I'm just trying to understand how it works:

The paper mentions that the interaction-level normalization factors are computed by inverse of Brownian + Technical noise terms: 1/(B+T).
When I'm looking at the code it seems like B+T is being used and not the inverse of it (lines 898, 1585,.. in chicdiff.R). Is that because since it is given to DESeq2 as a normalization factor, DESeq2 will use it as the denominator?

Thanks!

worchard commented 4 years ago

Hi Armen,

Thank you for pointing this out, I think there is a mistake in the supplementary material of the manuscript! You're right that DESeq2 will use it as a denominator, so in Chicdiff we combine the B + T (not 1/(B + T)!) with the DESeq sample-level size factor. This combined normalisation factor is then used as the denominator in the DESeq2 part of Chicdiff to calculate the normalised read count for each interaction in each sample. Will need to fix this :)