Using this model to trinarize gene expression, we call a gene expressed when P(Θ > f) > (1 - PEP), where f is the population fraction of cells expressing the gene, and PEP is the desired posterior error probability (also called local false discovery rate, or FDR). For example, with PEP = 0.05, there is less than 5% risk, given the observations, that the expressed call is wrong. Similarly, we call a gene not expressed when P(Θ > f) < PEP. For values between 1-PEP and PEP, we call the gene indeterminate. Note that PEP is applied individually to each gene (hence, ‘local FDR’) and the actual genome-wide FDR will be strictly equal to or lower than PEP.
