How to merge heterogeneously normalized RNA-seq datasets for meta-analysis?

Question

[pipelinfections]

Hi everyone,

I'm working on a meta-analysis involving 60< RNA-seq and array (NanoString) datasets with normalized gene expression values. However, the normalization methods vary: some use TPM, others FPKM or even CPM.

Currently, I'm merging all these datasets into a single expression matrix and applying quantile normalization on the merged matrix before performing DGE analysis.

Is this approach scientifically sound, or should I consider a different approach for integrating these datasets?

I also have access to the FASTQ files, so I could reprocess the data from scratch if that is the more appropriate approach.

Thanks for your help in advance!

[ATpoint]

Isn't the point of meta-analysis to basically analyze each dataset, due to heterogeneity across datasets, individually and then combine results into a meta-ranking to reveal the consistent, therefore reproducible, trends? I would never combine such heterogeneous data into a single quantitative analysis.

[pipelinfections]

Could you please elaborate on why you would never combine them into a single table? There are several papers that do this, e.g. this one, which states: "Gene expression data from all eligible datasets were combined into a single table, quantile normalized and scaled to 1000."

[ATpoint]

Because data are heterogeneous, on entirely different scales, assay (vastly) different amounts of genes and are inherently incomparable quantitatively. Just because there is a paper doing something does not mean its good or meaningful. That is my take on it, you are free to disagree.