merge my own RNA-seq data with TCGA UCSCXena data sets. and proper normalisation

Question

[theodore]

Hello all, and thank you in advance for your input.

I was reading this publication where they used TCGA data from the TCGA concortium and I would like to merge in the dataset RNA-seq data from patients n=5 of another type of cancer. One data set has featurecounts counts and the other one RSEM, as in the TCGA. The questions are

1. I assume that using different count program it will not affect very dramatically the data, well there are publications that claim this or the other, I am unsure.

2. Regarding normalisation I would assume that before merging my own data set I would need to do something like this:

    featureCounts_table <- read.table(paste0(dir.in, "geneCounts.txt"),head = T, sep = "\t",skip = 1, row.names = "Geneid")
    gene.length <- featureCounts_table$Length
    convertCounts(my_countMatrix+1, "TPM",  gene.length, log) #from DGEobj.utils cran package

1. How does TCGA Xena handles batch effects? Is it relevent if I only use the data for the GSVA package (single sample GSEA approach)

this is how I am getting the data from XENATCGA:

xe <- XenaGenerate(subset = XenaHostNames == "tcgaHub")
xe %>% XenaFilter(filterDatasets = "clinical") -> xe_clinical
xe %>% XenaFilter(filterDatasets = "HiSeqV2_PANCAN$") -> xe_rna_pancan
xe_clinical.query <- XenaQuery(xe_clinical)
xe_clinical.download <- XenaDownload(xe_clinical.query,  destdir = "UCSC_Xena/TCGA/Clinical", trans_slash = TRUE, force = TRUE)
xe_rna_pancan.query <- XenaQuery(xe_rna_pancan)
xe_rna_pancan.download <- XenaDownload(xe_rna_pancan.query,  destdir = "UCSC_Xena/TCGA/RNAseq_Pancan", trans_slash = TRUE)

As a note I found the following statments: "For the PANCAN gene expression dataset, we combined all the data from all the TCGA cohorts. This data is mean normalized across the entire cohort in the visualization (https://genome-cancer.ucsc.edu/proj/site/help/#Normalize_columns). Each individual cohort was Level_3 Data (file names: *.rsem.genes.normalized_results), all of which were downloaded from TCGA DCC and then were log2(x+1) transformed."

[Radu Tanasa]

You can analyze your data independently and then do a meta-analysis using both (TCGA and yours). At that point you can look at pathway overlap, etc. As long as your metadata is consistent, you can draw some useful conclusions.

[theodore]

Thank you for your answer, just to clarify, you recommend to loop the gene expression matrix, per samples, and parse it to GSVA idependetly and then merge the data/the GSVA geneset scores to start doing some meta analyis.

Answer 1

[Zhenyu Zhang]

For a careful analysis, how about you download data from the GDC, and run your samples via the GDC RNA-Seq pipeline. Then you can use the started DESeq/EdgeR/Limma analysis control for covariates.

[theodore]

the whole RNAseq pipeline from fastq to counts of all those thousand samples?

[Zhenyu Zhang]

I don't know how many samples you have. TCGA samples are already called, so that you only need to call your own samples.

[theodore]

OK, that makes sense. I will try that. Thank you