Speaker
Description
Single-cell omics methods promise to revolutionize our understanding of gene regulatory processes, offering genome-wide measurements at single-cell resolution. However, there are three central challenges: they are high-dimensional, noisy, and provide only snapshots.
Together, these challenges obstruct the extraction of accurate time-resolved gene expression dynamics. To still explore these datasets, the field has relied on embeddings like t-SNE or UMAP, even though such methods distort the intrinsic structure of the data.
To address this, we developed a broadly applicable Bayesian method called Bonsai, which reconstructs the most likely tree that relates any set of high-dimensional objects while rigorously accounting for heterogeneous measurement noise. We find that Bonsai exploits a “blessing of dimensionality”: representing cell-to-cell relations with a tree structure becomes virtually perfect at high dimensionality.
Although Bonsai can be used to explore data from a broad range of fields, it is particularly natural for studying differentiation processes, where all cells are related through a cell-division tree. We therefore applied Bonsai to a blood cell dataset where it recapitulated known differentiation trajectories based on only one snapshot of scRNA-seq data. Moreover, Bonsai finds convincing evidence that NK cells can derive from both myeloid and lymphoid ancestors in vivo, a novel and biologically significant result.
