Using TCGA RSEM data for isoform analysis

[cathp...@gmail.com]

Hi, I'm struggling to get my head around a few aspects of the publicly available TCGA RNA-seq data and what the best way to use the data is.

I'd like to compare the expression of 4 transcript variants of the same gene both within samples and between different samples (for example, tumours from different tissues or normal vs tumour samples).

I understand that TPM is the best way to analyse the relative abundance of each isoform within the same sample.

I've also read on a different post in this group that to compare abundances across different samples then a normalization method such as TMM needs to be applied.

My problem is that TCGA only provides the 'expected count' and TPM (scaled estimate x 10^6) values and not the FPKM values. For comparing isoform abundance across many samples (while also correcting for transcript length) I think that from what I've read the best way would be to normalize the FPKM values using TMM - is this right? I could also use the 'normalized' isoform data provided by TCGA but my understanding is that this is not corrected for transcript length.

So I guess what I'm asking is how to get transcript abundances that are both comparable between samples and corrected for transcript length given the somewhat limited output data provided by TCGA:

• Is it possible to correct the quantile normalized expected counts for transcript length? Where do I find the correct transcript lengths to use?
• Or alternatively is there a way to convert the TPM values back to FPKM values with just the data provided by TCGA? (expected count and 'scaled estimate', which I think might also be called tau value)
• Can the TMM normalization method be applied to TPM values? This doesn't sit quite right somehow but I'm wondering on the validity.

I apologise for my very limited understanding on this topic, I'm a bit overwhelmed and seem to be just getting more confused the more I read!

Thanks so much for your help.

Catherine