search

zhangyuqing / ComBat-seq

Batch effect adjustment based on negative binomial regression for RNA sequencing count data
154 stars 39 forks source link

Working with normalized datasets #2

Closed tengfei-emory closed 4 years ago

tengfei-emory commented 4 years ago

I suppose that we should input the raw count matrix instead of normalized count matrix to ComBat-seq so that the negative binomial assumption would satisfy. However, sometimes we may only obtain a normalized dataset, without raw count matrix. Are there any suggested strategies about how to apply ComBat-seq for that scenario? Thank you.

zhangyuqing commented 4 years ago

Hi @tengfei-emory, great question! Strictly speaking, ComBat-Seq should not be used in that scenario (and neither should edgeR or DESeq2), since the normalized data violate the assumptions under these methods.

There's a thread on Bioconductor about using normalized counts in edgeR (https://support.bioconductor.org/p/65890/#65910), which I find similar to our discussion.

tengfei-emory commented 4 years ago

Thank you so much for the information!