Variance Partitioning Analysis

Variance partitioning analysis on each gene

Usage

fitVarPart(
  x,
  x.null = NULL,
  distr.method = c("trigamma", "lognormal"),
  lambda.method = c("parametric", "mean"),
  ...
)

# S4 method for class 'lucida'
fitVarPart(
  x,
  x.null = NULL,
  distr.method = c("trigamma", "lognormal"),
  lambda.method = c("parametric", "mean"),
  ...
)

Arguments

x

model fit with lucida()

x.null

null model fit with lucida()

distr.method

use either the "lognormal" or "trigamma" formulas from Nakagawa, et al. (2017)

lambda.method

use either "parametric" or "mean" method to estimate the mean rate for count models

...

other args

Details

Using lambda.method = "parametric" requires refitting the model without the fixed effects. This is more accurate, especially for responses with low counts. Using lambda.method = "mean" is a fast approximation that works well for responses with sufficient counts. This methid can _under-estimate_ the count noise.

See also

BatchRegression::varpart(), fastglmm::varpart()

Examples

library(SingleCellExperiment)

# Load example data
data(example_sce, package="muscat")
sce <- example_sce

# Compute library size for each cell
sce$libSize <- colSums(counts(sce))

# Specify regression formula and cell annotation 
form <- ~ group_id + (1|sample_id)
fit <- lucida(sce, form, "cluster_id", verbose=FALSE)
#> B cells 
#> CD14+ Monocytes 
#> CD4 T cells 
#> CD8 T cells 
#> FCGR3A+ Monocytes 

# Model with only intercept and random effect
form <- ~ (1|sample_id)
fit.null <- lucida(sce, form, "cluster_id", verbose=FALSE)
#> B cells 
#> CD14+ Monocytes 
#> CD4 T cells 
#> CD8 T cells 
#> FCGR3A+ Monocytes 

# Variance partitioning analysis
vp <- fitVarPart(fit, fit.null)

plotVarPart(sortCols(vp))