(the subtypes for which we had the lowest sample sizes) (p < 0.00001 for uniform null, and p < 10−12 for simulated null). We corrected the binomial test p-values using the Benjamini-Hochberg procedure. To verify biological reproducibility, and because defining an archetype and computing discontinuities from the same data can lead to inflated p-value distributions,84 we calculated the archetypes using data from monkey P and computed the discontinuities on data from monkey F. To project the archetypes learned from monkey P onto monkey F, we multiplied the normalized expression matrices of monkey F by the pseudoinverse of the archetype vectors. Using the DSN algorithm we found several archetypes of particular interest including an archetype representing CNR1 signal in hybrid cells, caudate signal in D1- and D2- MSNs, core signal in D1- and D2- MSNs and TAC3 signal in shell D1- MSNs. To provide molecular markers for these archetypes we calculated each gene's Pearson correlation with each archetype of interest (Data S2 and S3; Tables S2 and S3).
