The denominator of this expression can be interpreted as a pseudo-reference sample obtained by taking the geometric mean across samples. Thus, each size factor estimate sˆj is computed as the median of the ratios of the j-th sample's counts to those of the pseudo-reference.

But what if there are zero counts? The geometric mean will appear to be zero. And there is no devision by 0. Who does it handle genes with zero counts in some condition? Or the assumption is that if there are no reads on one condition, then the whole gene should be excluded?