cluster_velo_txis: Simple Louvain, density, or k-means clustering. Doesn't use...

In trichelab/velocessor: The world's dumbest single-cell RNA velocity calculator and loader

Description

If a reduced dimension representation named "HARMONY" exists, the graph will be constructed on that, otherwise "PCA" will be used (or added if not found) unless use.dimred is explicitly specified by the user or "how" == "density". For mbkmeans, k is the number of clusters and "HARMONY" is used if found. For densityClust, the default representation to cluster upon is "UMAP".

Usage

cluster_velo_txis(
  txis,
  k = 20,
  type = "rank",
  use = NULL,
  how = c("louvain", "density", "mbkmeans"),
  ret = c("sce", "clust"),
  ...
)

Arguments

txis	a SingleCellExperiment
k	number of nearest neighbors, or clusters for mbkmeans (20)
type	what type of SNNGraph to construct ("rank")
use	which reducedDimension embedding to use (autodetermine)
how	cluster via "louvain" (default), "density", or "mbkmeans"?
ret	return "clust" assignments or (default) an "sce"?
...	optional arguments to pass to clustering function(s)

Details

It is not yet clear whether an approach like that taken in Rphenoannoy would allow for faster and therefore bootstrap-able clustering, and this is something that will be of interest going forwards. In some cases, using rank or jaccard distances with Louvain clustering produces overly fine- grained clusters, to the extent that something like SPADE may be needed to reconcile the results. The 'phenograph' algorithm is itself just Louvain clustering on 'jaccard' distance ("type"), something that 'rank' improves.

The default number of PCAs or Harmony components "k" is set equal to that which Seurat uses by default, not for any particularly good reason. Note that k means something different (cluster number K) for minibatch K-means, but we leave the default the same (because YOLO, and you can always merge). (Which is to say, also not for any particularly good reason.)

It would probably be a good idea to assess stability with bluster here.

Value

       Cluster labels (if ret=="clust") or a now-labeled "sce"

References

Louvain clustering: Blondel VD, Guillaume JL, Lambiotte R, Lefebvre E (2008). Fast unfolding of communities in large networks. Journal of Statistical Mechanics: Theory and Experiment. 2008 (10): P10008. doi:10.1088/1742-5468/2008/10/P10008 \ Density clustering: Rodriguez A, Laio A. (2014). Clustering by fast search and find of density peaks. Science, 344(6191), 1492-1496. doi:10.1126/science.1242072 \ Minibatch K-means: Sculley D. (2010). Web-Scale K-Means Clustering. WWW 2010, Raleigh, North Carolina, USA, ACM 978-1-60558-799-8/10/04. https://doi.org/10.1145/1772690.1772862 \ Hicks SC, Liu R, Ni Y, Purdom E, Risso D. (2020). mbkmeans: fast clustering for single cell data using mini-batch k-means. bioRxiv 2020.05.27.119438 doi: https://doi.org/10.1101/2020.05.27.119438

trichelab/velocessor documentation built on Jan. 5, 2022, 6:27 p.m.