We therefore started decomposing RNA-seq datasets into program activity (wrote a small algorithm for that). When we looked at the AML atlas data, we lived a big surprise: One of our Perturb-seq programs predicts long-term survival in AML very well! Independent of genotype. (3/n)
