TPM biological replicates, geometric mean

Question

[joha]

Hello, I have a TPM matrix from my RNA-seq experiment across several tissues, and I need to calculate Tau specificity index, which requires "merging" biological replicates into one single "expresion" by gene by tissue. I decided to calculate the geometric mean for these biological replicates (on Not log transformed TPM), so I previously replaced zero values by the minimal non-zero value (across entire matrix) which I found to be 0.0003. I prefered to do all this on non-log TPM as i need it also for other kind of visual reconstructions , in any case I think I can logTranform those geom_meanTPMs for Tau calculation. Is this approach logical (is my zero replacement strategy acceptable for calculating the geom mean?)?.