FitRegularizedClassificationGlm: FitRegularizedClassificationGlm

In bimberlabinternal/CellMembrane: A package with high-level wrappers and pipelines for single-cell RNA-seq tools

FitRegularizedClassificationGlm

Description

Treating gene expression like a classification problem, this function trains a penalized model to classify a metadata feature.

Usage

FitRegularizedClassificationGlm(
  seuratObj,
  metadataVariableForClassification = NULL,
  rescale = TRUE,
  numberOfVariableFeatures = 3000,
  assay = "RNA",
  layer = "scale.data",
  devianceCutoff = 0.8,
  split = NULL,
  returnModelAndSplits = F
)

Arguments

seuratObj	a Seurat object
metadataVariableForClassification	The metadata feature to be classified. If non-binary, then multinomial regression will automatically be performed.
rescale	The feature selection will optimize for "heatmap-interpretable genes" so the features are intended to be scaled. If TRUE, this will rescale the variable features.
numberOfVariableFeatures	A parameter to select how many features should be selected as variable for scaling, by default, all genes will be used.
assay	Seurat Object's assay
layer	layer within the Seurat object assay. Recommended to be "scale.data".
devianceCutoff	Tolerance for model error when deciding how much regularization should be performed. 1 = no tolerance for error, 0 = intercept only, no genes used for prediction.
split	the option to provide a previous model's training/testing set. This is necessary if you're performing multiple iterations of model fitting.
returnModelAndSplits	A boolean option to return a list containing the fitted model and training/testing splits in addition to the useful features.

Value

A vector of genes useful for classification and, optionally, the model and training/testing sets.

bimberlabinternal/CellMembrane documentation built on Dec. 20, 2024, 11:13 p.m.