1
$\begingroup$

I'm trying to download data from the TCGA for gene expression analyses in R, but I'm in doubt if I should use FPKM, FPKM-UQ or counts? When the dataset is in counts, I suppose it's raw data, isn't it? So what's the best unit to compare multiple datasets?

I'm planning to use limma or Dseq2 for GE analysis and found that with Dseq2 I need to use count(non-normalized???) data... is that correct? so what's the best package and working strategy?

thank you very much, Fabiano

$\endgroup$
3
$\begingroup$

Type of data you need depends on the downstream applications and since you would like to carry out DEA with DESeq2, you would need raw counts (non-normalized).

There are many ways to import/download TCGA data, one such tool, the TCGAbiolinks package gives a nice interface for not only downloading the read count (or pre-processed) data but also associated clinical data.

$\endgroup$
9
  • $\begingroup$ yes, I'm using TCGAbiolinks to download the data... what's the difference between limma and DESeq2? many thanks, @haci $\endgroup$ Aug 31 '20 at 15:13
  • $\begingroup$ I'm planning to compare my gene of interest in different types of cancer with normal tissue.... How many datasets of each kind of cancer and normal tissue should I include? $\endgroup$ Aug 31 '20 at 15:21
  • $\begingroup$ limma was designed for microarray data, but can be used for RNA-seq by using the voom transformation. However, the gold standard these days is DESeq2. As for how many datasets you need... what's your hypothesis? What are you looking for? Do you just want to blindly do tumor vs normal for multiple cancer types and check what comes up? $\endgroup$
    – csgroen
    Sep 4 '20 at 9:44
  • $\begingroup$ Hi @csgroen. yes, I just want to investigate my gene of interest in different types of cancer. Maybe I don't even need an extensive differential expression analysis. can I just compare my gene of interest in cancer cells with normal tissue using t student? it is probably enough to answer my question.. $\endgroup$ Sep 4 '20 at 20:40
  • $\begingroup$ If it is just for gene of interest, you can maybe do it using XenaBrowser... easier than downloading all the data using TCGAbiolinks then doing differential analysis. $\endgroup$
    – csgroen
    Sep 4 '20 at 20:48
1
$\begingroup$

You should use "counts" from the list of gene expression metrics that you provided rather than either FPKM or FPKM-UQ. FPKM is a normalized gene expression metric and an acronym for fragments per kilobase per million mapped reads. FPKM normalizes for both sequencing depth and genome size. Those specific packages want non-normalized counts as the expression metric. This means the number of reads mapped to a particular gene or feature. This is a slightly ambiguous metric but it can be the number of single-end reads mapped to a loci, the amount of pair end reads counted towards a loci, or the amount of super-reads (the theoretical sequence that spans the pair of paired-end reads which includes the presumed sequence between the first and second strand of a set of pair end reads) counted toward a loci. These values should be integers based on the assumptions of RNA-Seq and microarray technology which measures RNA transcripts present in a sample and precludes transcripts from spanning multiple genes. It does not matter which of these metrics you use as long as you are consistent. Those Bioconductor packages provide several different normalization methods but rely on read counts as their input.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.