Milo performs single-cell differential abundance testing. Cell states are modelled as representative neighbourhoods on a nearest neighbour graph. Hypothesis testing is performed using either a negative bionomial generalized linear model or negative binomial generalized linear mixed model.