Our analysis across multiple cell types revealed that some perturbations yielded universal signatures, whereas 43% of compounds yielded cell-type selective gene expression signatures. The fact that many compounds yield a universal signature regardless of cell type also has important implications. Specifically, the value of continuing to profile such compounds across a large number of cell lines is probably low. Future iterations of the CMap might therefore benefit from an adaptive experimental design whereby the selection of future cell lines is chosen based on the performance of an initial set.
