How do we handle TPM values after converting from Ensembl to HUGO genes?

Question

[Jordan]

Hi,

I have generated TPM values using TPMCalculator on human samples, using ensembl genome. Now, I have a matrix of TPM values with rows as ensembl genes and columns as samples.

I converted the ensembl genes to HUGO genes and lost a lot of rows, as they do not have corresponding HUGO gene id's.

I wanted to know how we can handle TPM values now? TPM values usually sum up to million in a sample, but after conversion and losing few genes, the sum reduces to about half a million.

Is it ok to use this new TPM matrix which has fewer genes for further analysis? If not, what are the possible solutions?

Thanks for the help!

[jared.andrews07]

What does your "further analysis" entail? This is a common issue, and where possible, it's usually best to retain the original gene IDs somehow and use the gene symbols for labelling/an additional column downstream.

[Jordan]

It involves running immunedeconv and unfortunately it needs HUGO id's.

Answer 1

[Istvan Albert]

A TPM is defined over the entire transcript as it existed at the time of sequencing - normally it should not be changed based on how many rows you keep, the standard protocol should be to keep the values as they are.

That being said if you load the values into another analysis step that does expect the numbers to add up to something and that makes use of that information in some manner then you should rescale them. Their relative sizes won't change after scaling.

Answer 2

[Mensur Dlakic]

Not sure that you need any conversion because TPM is just a number that works for an arbitrarily chosen scale. You might as well simply adopt a TPHM scale (transcripts per half-million).

If you really want to do the conversion, I suggest you work your way back from TPM to counts before removing samples without HUGO IDs.