the subtypes’ annotations and designated those archetypes as the transition axes between the subtype pairs. To determine if the transitions were discrete, we used regression discontinuity design (RDD).83 The null hypothesis of this test is that the gene expression is, solely, a linear function of the transition archetype weights, and adding a threshold does not add additional information. Therefore, we computed p-value distributions across all correlated genes for each connected pair. We then tested whether the number of genes that had significant discontinuities (p < 0 .05, RDD) was greater than expected by chance using a binomial test. We performed the binomial test with two different null distributions, a uniform null distribution, and a simulated null distribution where we randomly shuffled the cell labels and performed the RDD 100 times per gene. Regardless of which null distribution we used, the binomial tests indicated more statistically significant discontinuities than predicted by chance for every subtype pair except for D1S and D2S (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
