Single cell transcriptomic data are being ammassed by many laboratories and are revealing an amazing and sometimes overwhelming degree of heterogeneity within organisms, tissues, and even within each individual cell type. The cell similarity graph or network is a mathematical object at the core of such data sets, encoding phenotypic heterogeneity in a simplified yet powerful form. I will give an overview of my lab operations, centered around cell graphs and aimed at generating sound and interesting hypotheses for biomedicine via data exploration and deep experimental collaborations. I will present northstar, a new cell clustering/classification approach that is particularly well suited for cancer and developmental biology. I will then discuss two of our recent adventures in biomedicine: (1) constructing a cell atlas of the neonatal lung and (2) understanding the corruption of hematopoietic gene regulatory networks during acute myeloid leukemia.