Our procedure for selecting Landmark Genes was data-driven and the simulations presented above indicate that both the landmark and inferred genes capture relevant information about cell state. However, given a new state, any inference algorithm will only work if a fair number of the landmark genes are expressed in that state. We examined expression across lineage using the Genotype Tissue Expression (GTEx) RNA-seq dataset (DSGTEx-RNA-seq) of 3,176 patient-derived expression profiles from 30 different tissue types (Figure S1B). We quantified the expression levels of the landmark genes reported in the dataset and observed that at a RPKM threshold of 1 at least 86% of Landmark Genes are expressed in each of the 3,176 samples (with an average of 92% expressed in each sample), and that range of expression is similar across tissue types.
