Find variable features fitting LOESS

Find highly variable features by fitting a locally polynomial regression model to the log(mean) and log(variance) of downsampled features.

FitMeanVar(object, ...)

# S3 method for class 'Seurat'
FitMeanVar(
  object,
  assay = NULL,
  layer = NULL,
  nfeatures = 20000,
  loess.span = 0.1,
  min.cutoff = 10,
  weight.mean = 0,
  bins = 1000,
  sample_per_bin = 50,
  key = "dsLoess",
  verbose = FALSE,
  ...
)

# S3 method for class 'StdAssay'
FitMeanVar(
  object,
  layer = NULL,
  loess.span = 0.1,
  nfeatures = 20000,
  min.cutoff = 10,
  weight.mean = 0,
  bins = 1000,
  sample_per_bin = 50,
  key = "dsLoess",
  verbose = FALSE,
  ...
)

# S3 method for class 'Assay'
FitMeanVar(
  object,
  layer = NULL,
  loess.span = 0.1,
  nfeatures = 20000,
  min.cutoff = 10,
  weight.mean = 0,
  bins = 1000,
  sample_per_bin = 50,
  key = "dsLoess",
  verbose = FALSE,
  ...
)

# Default S3 method
FitMeanVar(
  object,
  nfeatures = 20000,
  loess.span = 0.1,
  min.cutoff = 10,
  weight.mean = 0,
  bins = 1000,
  sample_per_bin = 50,
  random.seed = 1234,
  verbose = FALSE,
  ...
)

# S3 method for class 'data.frame'
FitMeanVar(
  object,
  loess.span = 0.1,
  weight.mean = 0,
  bins = 1000,
  sample_per_bin = 50,
  random.seed = 1234,
  verbose = FALSE,
  ...
)

Arguments

object

A Seurat object

...

Arguments passed to other methods

assay

Name of assay to use. If NULL, use the default assay.

layer

Name of layer to use. If NULL, use the default layer(s).

nfeatures

Number of features to selected as top variable features.

loess.span

span parameter passed to the stats::loess() function

min.cutoff

Minimum number of counts for a feature to be eligible for variable feature selection.

weight.mean

How much to weight the ranking of features according to their mean. Setting weight.mean=0 will rank features according to their residual variance only.

bins

Number of bins to use when downsampling features across the range of mean count values.

sample_per_bin

Number of features to select per mean count bin in feature downsampling step.

key

Key to use when storing the highly variable feature information in the assay.

verbose

Display messages.

random.seed

Random seed to set for sampling.

Value

Returns a SeuratObject::Seurat() object

Details

This function is similar to the Seurat::FindVariableFeatures() function (with selection.method="vst"), but downsamples the features evenly across the range of mean values. This speeds up fitting of the loess curve when the number of features is large.

The function also provides the ability to combine ranking of features according to their mean count and their residual variance, using a weighted rank sum with weights set by the weight.mean parameter. This can help to avoid selecting features with high residual variance but very low mean.

Examples

FitMeanVar(atac_small)
#> An object of class Seurat 
#> 1323 features across 100 samples within 3 assays 
#> Active assay: peaks (323 features, 323 variable features)
#>  2 layers present: counts, data
#>  2 other assays present: bins, RNA
#>  2 dimensional reductions calculated: lsi, umap
FitMeanVar(object = atac_small[["peaks"]])
#> ChromatinAssay data with 323 features for 100 cells
#> Variable features: 323 
#> Genome: hg19 
#> Annotation present: TRUE 
#> Motifs present: TRUE 
#> Fragment files: 0