qTune: Tuning the choice of q (number of clusters) before running...

In BayesSpace: Clustering and Resolution Enhancement of Spatial Transcriptomes

Description

Before running spatialCluster(), we recommend tuning the choice of q by choosing the q that maximizes the model's negative log likelihood over early iterations. qTune() computes the average negative log likelihood for a range of q values over iterations 100:1000, and qPlot() displays the results.

Usage

qPlot(sce, qs = seq(3, 7), force.retune = FALSE, ...)

qTune(sce, qs = seq(3, 7), burn.in = 100, nrep = 1000, ...)

Arguments

sce	A SingleCellExperiment object containing the spatial data.
qs	The values of q to evaluate.
force.retune	If specified, existing tuning values in sce will be overwritten.
...	Other parameters are passed to spatialCluster().
burn.in, nrep	Integers specifying the range of repetitions to compute.

Details

qTune() takes the same parameters as spatialCluster() and will run the MCMC clustering algorithm up to nrep iterations for each value of q. The first burn.in iterations are discarded as burn-in and the log likelihood is averaged over the remaining iterations.

qPlot() plots the computed negative log likelihoods as a function of q. If qTune() was run previously, i.e. there exists an attribute of sce named "q.logliks", the pre-computed results are displayed. Otherwise, or if force.retune is specified, qplot() will automatically run qTune() before plotting (and can take the same parameters as spatialCluster().

Value

qTune() returns a modified sce with tuning log likelihoods stored as an attribute named "q.logliks".

qPlot() returns a ggplot object.

Examples

set.seed(149)
sce <- exampleSCE()
sce <- qTune(sce, seq(3, 7), burn.in=10, nrep=100)
qPlot(sce)

BayesSpace documentation built on Nov. 8, 2020, 8:03 p.m.