Is possible to recreate RNASeq count dataset from a dataset normalized with FPKM ?

Question

[elmahy2005]

Given a dataset of RNA-Seq expression values normalized with the FPKM method, Is it possible to restore the original count dataset or create a new dataset that behaves very similar to the original count matrix (i.e. we can use in Poisson distribution based models)?

[Benn]

Unfortunately, it is not possible to calculate raw counts from RPKM data. Best is to start with bam files, and use a program such as featureCounts to generate raw counts.

[JC]

Agree on this, because after normalization who knows how the original values were modified.

Answer 1

[h.mon]

If you have the library sizes and effective transcript lengths, you can calculate the original counts. If you have the FPKMs alone, you can't. See the formula for FPKM (from What the FPKM? A review of RNA-Seq expression units):