The cells of multicellular organisms can be hierarchically arranged based on their capacity to generate other cell types. This “potency”—ranging from totipotent to differentiated—declines with each step of that hierarchy. From the observation that the diversity of genes expressed by cells similarly declines with differentiation, researchers led by Ludwig Stanford’s Aaron Newman developed in 2020 a computational framework named CytoTRACE to predict the potency of cells based on single-cell RNA sequencing data. In October, Aaron and his colleagues reported in Nature Methods their development of CytoTRACE 2, an AI-powered improvement on its predecessor. For CytoTRACE 2, the researchers trained a machine learning algorithm on huge datasets of gene activity known to be associated with various levels of differentiation. They then sorted the cells into six levels of potency and 24 subcategories. While the original CytoTRACE can only compare the potency of cells within a single dataset, CytoTRACE 2 can do so across datasets, assigning an absolute measure of potency to any given cell across tissues. Most notably, its results are interpretable—providing insight into such things as the gene expression programs that determine potency. Illustrating its utility for cancer research, the researchers showed CytoTRACE 2’s predictions correspond to established stem cell signatures in a leukemia and could identify the known multilineage potential of oligodendroglioma cells. These results open the door to more precise identification of high potency cells
in cancer.
Improved reconstruction of single-cell developmental potential with CytoTRACE 2
Nature Methods, 2025 October 27