Chapter 16 Trajectory Analysis with ArchR
To order cells in pseudo-time, ArchR creates cellular trajectories that order cells across a lower N-dimensional subspace within an ArchRProject. Previously, we have performed this ordering in the 2-dimensional UMAP subspace but ArchR has improved upon this methodology to enable alignment within an N-dimensional subspace (i.e. LSI). First, ArchR requires a user-defined trajectory backbone that provides a rough ordering of cell groups/clusters. For example, given user-determined cluster identities, one might provide the cluster IDs for a stem cell cluster, then a progenitor cell cluster, and then a differentiated cell cluster that correspond to a known or presumed biologically relevant cellular trajectory (i.e. providing the cluster IDs for HSC, to GMP, to Monocyte). Next, for each cluster, ArchR calculates the mean coordinates for each cell group/cluster in N-dimensions and retains cells whose Euclidean distance to those mean coordinates is in the top 5% of all cells. Next, ArchR computes the distance for each cell from cluster i to the mean coordinates of cluster i+1 along the trajectory and computes a pseudo-time vector based on these distances for each iteration of i. This allows ArchR to determine an N-dimensional coordinate and a pseudo-time value for each of the cells retained as part of the trajectory based on the Euclidean distance to the cell group/cluster mean coordinates. Next, ArchR fits a continuous trajectory to each N-dimensional coordinate based on the pseudo-time value using the
smooth.spline function. Then, ArchR aligns all cells to the trajectory based on their Euclidean distance to the nearest point along the manifold. ArchR then scales this alignment to 100 and stores this pseudo-time in the ArchRProject for downstream analyses.
ArchR can create matrices that convey pseudo-time trends across features stored within the Arrow files. For example, ArchR can analyze changes in TF deviations, gene scores, or integrated gene expression across pseudo-time to identify regulators or regulatory elements that are dynamic throughout the cellular trajectory. First, ArchR groups cells in small user-defined quantile increments (default = 1/100) across the cellular trajectory. ArchR then smooths this matrix per feature using a user-defined smoothing window (default = 9/100) using the
data.table::frollmean function. ArchR then returns this smoothed pseudo-time x feature matrix as a
SummarizedExperiment for downstream analyses. ArchR additionally can correlate two of these smoothed pseudo-time x feature matrices using name matching (i.e. positive regulators with chromVAR TF deviations and gene score/integration profiles) or by genomic position overlap methods (i.e. peak-to-gene linkages) using low-overlapping cellular aggregates as described in previous sections. Thus, ArchR facilitates integrative analyses across cellular trajectories, revealing correlated regulatory dynamics across multi-modal data.