Converting TCGA expression data from FPKM to TPM

Question

[Alex Reynolds]

For a given cancer type in the NIH Cancer Genome Atlas, I visit the data portal and download UNC RNASeqV2, level 3 expression data. Specifically, I grab files that end with the extension *.rsem.genes.normalized_results

Each file contains one line per gene, with the gene name and (I assume) its normalized FPKM expression value. I am assuming these data are normalized FPKM based on the filename and the UNC RNASeqV2 protocol description hosted on TCGA.

My questions are:

1. Are these expression data really measured in FPKM?
2. If they are, how should I convert from FPKM to TPM, for all the expression values for a given gene?

[gc]

You can't recover TPMs from gene-level FPKMs. The data on transcripts has already been lost.

[Michi]

I don't understand your comment. I've quickly compared the FPKMs for a given gene and it's transcripts and noticed (as one could expect) that the gene-level FPKM is the sum of all FPKM of it's transcripts. So it would not really make a difference if you calculate the TPM from gene or transcript-level FPKMs, I conclude. Hereafter one example:

genes.fpkm_tracking:ENSG00000196092    ENSG00000196092    PAX5    29.5427


isoforms.fpkm_tracking:ENST00000358127    ENSG00000196092    PAX5    5.41329
isoforms.fpkm_tracking:ENST00000520154    ENSG00000196092    PAX5    2.55302e-10
isoforms.fpkm_tracking:ENST00000523241    ENSG00000196092    PAX5    2.06415e-12
isoforms.fpkm_tracking:ENST00000377840    ENSG00000196092    PAX5    8.02239e-16
isoforms.fpkm_tracking:ENST00000377852    ENSG00000196092    PAX5    0.561218
isoforms.fpkm_tracking:ENST00000377853    ENSG00000196092    PAX5    5.90173e-10
isoforms.fpkm_tracking:ENST00000523145    ENSG00000196092    PAX5    2.63708e-10
isoforms.fpkm_tracking:ENST00000446742    ENSG00000196092    PAX5    0.387949
isoforms.fpkm_tracking:ENST00000520281    ENSG00000196092    PAX5    0.00123482
isoforms.fpkm_tracking:ENST00000377847    ENSG00000196092    PAX5    20.8611
isoforms.fpkm_tracking:ENST00000522003    ENSG00000196092    PAX5    0.474374
isoforms.fpkm_tracking:ENST00000414447    ENSG00000196092    PAX5    0.995077
isoforms.fpkm_tracking:ENST00000523493    ENSG00000196092    PAX5    0.663389
isoforms.fpkm_tracking:ENST00000524340    ENSG00000196092    PAX5    1.70926e-82
isoforms.fpkm_tracking:ENST00000522932    ENSG00000196092    PAX5    0
isoforms.fpkm_tracking:ENST00000520083    ENSG00000196092    PAX5    0.185101

[roy.granit]

Are you sure you need the TPM (Transcripts Per Million) data? If you are fine with the data at the gene level you should be OK as it is

[Alex Reynolds]

I'd like the TPM data, if possible.

[james.lloyd]

Quick question. If I have between sample normalized FPKMs, do I just sum the FPKMs of all the transcripts for a given gene within a sample, or do I sum all of those and for all those transcripts in the other samples. I'm just thinking, if you have three transcripts and two samples, that is different maths.

[Istvan Albert]

You need to post this as a new question and refer back to this thread if necessary. Each thread starts with a question followed by answers - new questions should not be posted in the answer section. That's what makes this site better than others. (Moderation: your answer will be moved to a comment)

Answer 1

[h.mon]

At the end of this blog post, a simple formula is provided to compute TPM from FPKM:

TPMi=( FPKMi / sum(FPKMj ) * 10^6

edit: well, from the protocol you linked, and also from this wiki, the UNC V2 RNA-Seq Workflow uses MapSplice+RSEM, so I guess measures are already given as TPM - check here and here.

[Alex Reynolds]

Thanks, the wiki link was a much better summary than what I had found previously.

Answer 2

[fabio-verdao]

Maybe it's a little bit old, but just for future access...

@h.mon answer your second question.

For your first question: 1. Are these expression data really measured in FPKM?

Following the wiki cited by @h.mon, *.rsem.genes.normalized_results as well as *.rsem.isoforms.normalized_results have measures in normalized_count (upper quartile normalized RSEM count estimates) and not RPKM, FPKM or TPM.

[masicong.smu]

Hello, so can I directly use the data in *.rsem.genes.normalized_results to do differential analysis, or pathway enrichment analysis, etc? Thanks!