Fig. 2

Workflow of the proposed simplicial complex model. We first reduce ambient dimension of the data via PCA to facilitate geometric analysis. We then fit simplices to each of the lineages using a shortest-path weight as an estimate of similarity between 2 points. We then use a novel objective function to robustly unmix each of the lineages. We reconcile shared sub-populations across lineages by training the noise based on several replicates of each of the robust unmixing runs. After a representative model has been derived from the replicates, we abstract the simplicial complex into a phylogeny, based on the dimensionality and connectivity of the simplices in the simplicial complex