Set two roots?

[josephryan]

Hello,

We've conducted 10x scRNA-Seq on 11 developmental timepoints (4-24 hours post fertilization) in animal where developmental fate of cells is set early (determinant cleavage). Our time course starts at 4 hours. Based on cell-lineage tracking and a t-SNE plot of all timepoints merged, we expect that each 4-hour cell has one of two distinct gene expression profiles. However, when we set 4 hours as the root state, we get a trajectory with a single root cluster (where we expect two root clusters). Is there a way to set 2 root states when running simulateRandomWalksFromTips?

If not, do you have any suggestions? I had a couple ideas that are both a bit painful and slightly wonky: (1) convert (coerce?) TPMs from an available bulk-RNA-seq dataset of pooled single-cell embryos into a set of UMIs, then read these into URD and try using them as a root, or (2) create a fake datapoint (pre-4hpf) that is a hybrid of the two real 4-hour clusters and set this extra timepoint as a root.

Any ideas would be welcome! Thank you!