Harmonized RNA-Seq Analysis of Different GDC Queried Projects

Question

[dk0319]

Are there any vignettes or resources anyone can recommend for performing RNA-seq in which different projects from GDC are compared?

In my specific case I am comparing unmatched metastatic prostate samples to prostate samples from a primary tumor, using harmonized star-counts.

I have tried TMM/voom/limma and when I performed PCA the majority of samples cluster based on sample type (eg. Met or Primary). However, this could easily be due to the projects being (Project 1=499/500 samples are Primary and Project 2=99/99 samples are Mets)

Ideally, I am interested in advice on how to best handle this sort of analysis along with advice on how to remove potential bias due to the samples coming from different projects.

Answer 1

[Zhenyu Zhang]

Batch effect correction is time consuming, and requires a lot of knowledge and investigations. I assume you can check those PanCan or PCWAG papers for some examples. Or you can play lazy and use automatic batch effects analysis (e.g. Combat).

A good resource for TCGA RNA-Seq batch effects can be found here https://bioinformatics.mdanderson.org/public-software/tcga-batch-effects/. You can also adopt their algorithm.