% Generated by roxygen2: do not edit by hand
% Please edit documentation in R/classifier.R
\name{train_classifier}
\alias{train_classifier}
\title{Train cell type classifier}
\usage{
train_classifier(
  train_obj,
  assay,
  slot = NULL,
  cell_type,
  marker_genes,
  tag_slot,
  parent_cell = NA_character_,
  parent_tag_slot = "predicted_cell_type",
  parent_classifier = NULL,
  path_to_models = "default",
  zscore = TRUE
)
}
\arguments{
\item{train_obj}{object that can be used for training the new model. 
\code{\link{Seurat}} object or \code{\link{SingleCellExperiment}} object
is supported.
If the training model has parent, parent_tag_slot may have been indicated. 
This field would have been filled out automatically 
if user precedently run classify_cells function. 
If no (predicted) cell type annotation provided, 
the function can be run if 1- parent_cell or 2- parent_classifier is provided.}

\item{assay}{name of assay to use in training object.}

\item{slot}{type of expression data to use in training object, omitted if 
train_obj is \code{\link{SingleCellExperiment}} object.}

\item{cell_type}{string indicating the name of the subtype
This must exactly match cell tag/label if cell tag/label is a string.}

\item{marker_genes}{list of marker genes used for the new training model}

\item{tag_slot}{string, name of slot in cell meta data 
indicating cell tag/label in the training object.
Strings indicating cell types are expected in this slot.
For \code{\link{Seurat}} object, default value is "active.ident".  
Expected values are string (A-Z, a-z, 0-9, no special character accepted) 
or binary/logical, 0/"no"/F/FALSE: not being new cell type, 
1/"yes"/T/TRUE: being new cell type.}

\item{parent_cell}{string indicated the name of the parent cell type, 
if parent cell type classifier has already been saved in model database.
Adjust path_to_models for exact database.}

\item{parent_tag_slot}{string, name of a slot in cell meta data 
indicating assigned/predicted cell type. Default is "predicted_cell_type". 
This slot would have been filled automatically 
if user have called classify_cells function.
The slot must contain only string values.}

\item{parent_classifier}{classification model for the parent cell type}

\item{path_to_models}{path to the folder containing the model database. 
As default, the pretrained models in the package will be used. 
If user has trained new models, indicate the folder containing the 
new_models.rda file.}

\item{zscore}{whether gene expression in train_obj is transformed to zscore}
}
\value{
\code{\link{scAnnotatR}} object
}
\description{
Train a classifier for a new cell type. 
If cell type has a parent, only available for \code{\link{scAnnotatR}}
object as parent cell classifying model.
}
\note{
Only one cell type is expected for each cell in object. 
Ambiguous cell type, such as: "T cells/NK cells/ILC", 
will be ignored from training.
Subtypes used in training model for parent cell types must be indicated
as parent cell type. For example, when training for B cells, 
plasma cells must be annotated as B cells in order to be used.
}
\examples{
# load small example dataset
data("tirosh_mel80_example")

# this dataset already contains pre-defined cell labels
table(Seurat::Idents(tirosh_mel80_example))

# define genes to use to classify this cell type (B cells in this example)
selected_marker_genes_B = c("CD19", "MS4A1", "CD79A")

# train the classifier, the "cell_type" argument must match 
# the cell labels in the data, except upper/lower case
set.seed(123)
classifier_b <- train_classifier(train_obj = tirosh_mel80_example,
assay = 'RNA', slot = 'counts', marker_genes = selected_marker_genes_B, 
cell_type = "b cells", tag_slot = 'active.ident')

# classify cell types using B cell classifier, 
# a test classifier process may be used before applying the classifier 
tirosh_mel80_example <- classify_cells(classify_obj = tirosh_mel80_example, 
classifiers = c(classifier_b), assay = 'RNA', slot = 'counts')

# tag all cells that are plasma cells (random example here)
tirosh_mel80_example[['plasma_cell_tag']] <- c(rep(1, 80), rep(0, 400))

# set new marker genes for the subtype
p_marker_genes = c("SDC1", "CD19", "CD79A")

# train the classifier, the "B cell" classifier is used as parent. 
# This means, only cells already classified as "B cells" will be evaluated.
# the "tag_slot" parameter tells the classifier to use this cell meta data
# for the training process.
set.seed(123)
plasma_classifier <- train_classifier(train_obj = tirosh_mel80_example, 
assay = 'RNA', slot = 'counts', cell_type = 'Plasma cell', 
marker_genes = p_marker_genes, tag_slot = 'plasma_cell_tag',
parent_classifier = classifier_b)

}