We split cells by major class according to the class assignment probability. For each cluster at level 1, we removed cells with conflicting classification (i.e., cells classified as Neuron in a cluster where the majority of cells are classified as Vascular). We performed the same analysis steps as for Level 1.