effective gene length and TPM values in Salmon's output files

[Tzachi]

Hi Rob,

I was trying to calculate the TPMs based on the values given for read counts and effective gene length in the quant.genes.sf and quant.sf files and encountered a problem with the gene level TPMs.
When I calculate the TPM value for transcripts I get identical results to the original Salmon value (I am doing it by dividing the number of reads by effective transcript length, then dividing by the sum of this ratio of all transcripts and then multiplying by a million).
However, when doing the same with the genes file - I get similar results to the original TPM values as given by Salmon, but not identical, and the variation between the original Salmon value and my calculated value seem to differ in their extent between genes.

How do you calculate the TPMs for genes? and how do you calculate effective gene length?

Sorry if I am missing something very basic here,

Thank you very much

Tzachi