Entering edit mode
10 months ago
star
▴
10
Hi! I'm very new to the bioinformatics scene and need advice on a project. One of the initial steps is to compare the deconvolution results of normal and tumor tissue RNA-seq (TPM) data from GTEx and TCGA (specifically ovarian cancer data), but I am confused about how to go about normalization and batch effect correction prior to doing this.
I am planning to use limma as mentioned by many papers I have found. However, I am still confused as to whether this is appropriate for my project, especially since I have seen posts online discouraging the use of TPM data.
How can I ensure that my normal and tumor tissue data are properly normalized / corrected for batch effects before running them through CIBERSORTx?
What your are proposing isn't trivial - you have to be very careful with batch correcting data from these disparate sources. I myself have done some work with skin GTEx data combined with TCGA melanoma data and there were virtually no normals in the TCGA data to facilitate batch correction, ultimately the data source was too confounded with sample type (i.e. cancer = TCGA and normal = GTEx) to make any reliable conclusions when comparing the two.
How many normals are in your ovarian cancer data? How do they cluster in PCA relative to the GTEx normals?
I suspects that not only they have different normalization method, the gene models are also different. Depends on the particular application, but if I am trying to do this carefully, I'd rather gather all the reads and redo STAR alignment and counting.
Yes, those details matter BUT thankfully you can get TGCA and GTEx gene counts that have been collected using the same pipeline from UCSC Xena https://xena.ucsc.edu/, hopefully OP is planning to start there at the very least