Pan-cancer RNA-seq data normalization

Question

[Brisket]

I have a set of FPKM data of several cancer cell lines and want to compare their gene expression.

Is it ok to transform them into logged numbers or TPM and calculate the fold change?

Or I have to calculate z-score?

[ATpoint]

Is this data from the same source or did you collect it from independent sources?

[Brisket]

Sorry for replying after so long. The data was from CellMiner, it contains 60 cell lines' RNA-seq FPKM data in a single excel sheet. I couldn't find the raw count datao I have to do my analysis based on FPKM.

[bkleiboeker]

Is it an option to download the reads and pseudo-align them yourself with kallisto/salmon? Most laptops could do this overnight, as long as you have room to store the .fastq files temporarily (account for approx 6*60 = 360GB of storage).

Answer 1

[bkleiboeker]

Typically it's not good practice to use FPKM (or other such values) for any sort of downstream statistical analyses, including differential expression analysis (read: calculating fold change).

Answer 2

[swbarnes2]

Mathematically, FPKM is not appropriate for comparison between samples. No amount of mathematical transformation will change that.