In kimberle9/rfca: Random Forest Cell Annotation of single-cell transcriptomics data

rfca

This package contains functions that trains a Random Forest Model with a labelled Seurat Object, for predicting cell types/states in unlabelled datasets. It also contains a pre-trained Random Forest model, as well as example datasets.

Introduction

Manual cell annotation of scRNAseq datasets, typically based on marker genes, can be time-consuming and biased. Being able to automatically predict cell types/states in a cell-by-cell and cluster-unbiased way is useful for fast and accurate phenotyping.

In addition, despite the increasing amounts of scRNAseq datasets being generated, thorough analysis of these datasets is lagging, and/or done in silos. This package comes with a preloaded Random Forest model based on different datasets and cell types/states, that will be constantly updated.

Installation

You can install this package from GitHub with devtools::install_github("kimberle9/rfca")

Manual

View the manual here: https://kimberle9.github.io/rfca/articles/vignette.html

Examples

Example 1: Creating a Seurat Object

# Load the rfca and Seurat libraries, as well as example datasets
library(rfca)
library(Seurat)

# Create Seurat Object with PCA and UMAP calculated
# Cell ranger raw data ("/filtered_feature_bc_matrix" directory) is required for this step
mySeuratObject <- createSeuratObjectPipeline(data.dir = "~/filtered_feature_bc_matrix", 
                                             nFeature_RNA_lower = 500, 
                                             nFeature_RNA_upper = 5000, 
                                             percent.mt = 5, 
                                             nfeatures = 2000, 
                                             dims = 20, 
                                             clusterResolution = 0.8)

# Assign cell type Idents to mySeuratObject manually, if you want to use it as a training dataset

Example 2: Using Random Forest to train and predict cell types

library(rfca)
library(Seurat)
data("exampleSeuratObjectUnlabelled")
data("exampleSeuratObjectLabelled")

# Create Random Forest Model with your labelled Seurat Object
myRandomForestModel <- createRFModel(exampleSeuratObjectLabelled)

# Create marker gene list from random forest model
markerGeneList <- createGeneLists(myRandomForestModel)

# Visualize Feature Plot based on marker gene list
myPlot <- cellMarkerPlots(exampleSeuratObjectLabelled, geneList = markerGeneList)
myPlot

# Predict cells based on your own Random Forest Model created above
autoLabelledSeuratObject <- predictCells(exampleSeuratObjectUnlabelled, myRandomForestModel)

# Predict cells based on my pre-loaded and pre-trained Random Forest Model
# With no model passed in, default model used is mouseBrain.
# See documentation for available tissue types. 
autoLabelledSeuratObject <- predictCells(exampleSeuratObjectUnlabelled)

# Visualize autoLabelledSeuratObject
DimPlot(autoLabelledSeuratObject)

Example 3: Defining microglia subtypes

library(rfca)
library(Seurat)

# Create Seurat object (microglia-only data)
microglia <- createSeuratObjectPipeline(data.dir = "~/filtered_feature_bc_matrix") 

# Define microglia subtypes based on cluster numbers
DimPlot(microglia)

# Build Random Forest Model with cluster number as labels
microgliaModel <- createRFModel(microglia)

# Phenotype subtype proportion for different mouse models
mouseModel1 <- predictCells(mouseModel1, microgliaModel)
mouseModel2 <- predictCells(mouseModel2, microgliaModel)
mouseModel3 <- predictCells(mouseModel3, microgliaModel)

kimberle9/rfca documentation built on Oct. 8, 2020, 5:44 a.m.