This is a test version of Biostars. For the public version, visit https://www.biostars.org.
Which data to choose for correlation examination

Hello I want to perform WGCNA or Spearman between multiple groups in TCGA pancancer data.

I am currently using Xena browser to download data and cBioPortal for mutation examination. I am undecided whether I should use "tcga_RSEM_gene_tpm" which is already TPM normalized, and contains 60k features but probably without batch correction and it will be difficult "to correct" with so many features.. or I should use "EB++AdjustPANCAN_IlluminaHiSeq_RNASeqV2.geneExp.xena" which is already after batch effect removal, however, contains much fewer features and will need an additional normalization.

I appreciate any help you can provide

correlation wgcna tpm batch spearman
• 1,700 views • link
updated by Zhenyu Zhang Zhenyu Zhang • written by lyan125 • lyan125

1 answer

I will use the batch-corrected data. I assume the second dataset you mentioned is the same as this one EBPlusPlusAdjustPANCAN_IlluminaHiSeq_RNASeqV2.geneExp.tsv in https://gdc.cancer.gov/about-data/publications/pancanatlas

Btw, I feel weird that now people like to download data from Xena and cBioPortal. How are you going to quote your data source and method if you do find something worth publishing? Why don't you get the data from the original PanCan publications where you have the full details on how data were generated.

• 0 views • link
written by Zhenyu Zhang Zhenyu Zhang

Thank you

Yes, of course, it's just very convenient and fast, much faster than downloading from publications or GDC itself It's good to save some time. As long as I acknowledge the way they did to get that kind of data, why shouldn't I use Xena or cBioPortal?

• 0 views • link
written by lyan125 • lyan125

Just a small note about this method. After performing multiple tasks, I would need to intersect the results (the genes) with 2 additional data frames, both of them TPM/FPKM normalized. If I understand correctly, this one can't be normalized to this mode unless I use raw counts with each transcript length to do so, am I correct?

• 0 views • link
written by lyan125 • lyan125

Yes, you need raw count data and gene model for TPM/FPKM

• 0 views • link
written by Zhenyu Zhang Zhenyu Zhang

It's all about reproducibility. What if after you publish your data, Xena and cBioPortal updates their data? I know there are many versions of TCGA expression data around. How could you or someone else reproducing your findings if the source data has been changed?

You don't have this issue if you quote data directly from a fixed publication, or use data from GDC where each analysis is versioned.

• 0 views • link
written by Zhenyu Zhang Zhenyu Zhang

Log in to answer this question.

More posts like this