R script to apply cell cycle ccAF classifier to single cell dataset.

ccAF.R

library(Seurat)
library(ranger)

# Modified function from Seurat V2
ClassifyCells = function (classifier, training.genes = NULL, training.classes = NULL, 
    new.data = NULL, ...) 
{
    features <- classifier$forest$independent.variable.names
    genes.to.add <- setdiff(x = features, y = rownames(x = new.data))
    data.to.add <- matrix(data = 0, nrow = length(x = genes.to.add), 
        ncol = ncol(x = new.data))
    rownames(x = data.to.add) <- genes.to.add
    new.data <- rbind(new.data, data.to.add)
    new.data <- new.data[features, ]
    new.data <- as.matrix(x = t(x = new.data))
    message("Running Classifier ...")
    prediction <- predict(classifier, new.data)
    new.classes <- prediction$predictions
    return(new.classes)
}

#############################################
### HEK293T cells from barnyard assay 10X ###
### http://cf.10xgenomics.com/samples/cell-exp/3.0.2/10k_hgmm_v3_nextgem/10k_hgmm_v3_nextgem_filtered_feature_bc_matrix.tar.gz ###
#############################################

setwd('/files/hgmm')
exp_mat = Read10X('filtered_feature_bc_matrix')

# Subset to only human genes
exp_mat = exp_mat[grep('hg19_',rownames(exp_mat)),] # Only human genes
rownames(exp_mat) = gsub('hg19_', '', rownames(exp_mat)) # Rename to remove hg19_
#exp_mat = exp_mat[,colnames(exp_mat)[which(colSums(as.matrix(exp_mat))>10000)]] # Only HEK293T cells

# Make Seruat object
HEK293T = CreateSeuratObject(counts=exp_mat)
HEK293T[["percent.mt"]] = PercentageFeatureSet(HEK293T, pattern = "^MT-")
pdf('HEK293T_vlnPlot.pdf', width=11, height=8.5)
VlnPlot(HEK293T, features = c("nFeature_RNA", "nCount_RNA", "percent.mt"), ncol = 3)
plot1 = FeatureScatter(HEK293T, feature1 = "nCount_RNA", feature2 = "percent.mt")
plot2 = FeatureScatter(HEK293T, feature1 = "nCount_RNA", feature2 = "nFeature_RNA")
CombinePlots(plots = list(plot1, plot2))
dev.off()

# Subset out good HEK293T cells  (ncol = 3468, nrow = 57,905)
HEK293T = subset(HEK293T, subset = nFeature_RNA > 4500 & nFeature_RNA < 100000 & percent.mt < 30)
dim(HEK293T)

# Find variable features, scale data, and do dimensionality reduction
HEK293T = NormalizeData(object = HEK293T)

# Load the classifier
ccAF = ('ccAF.rds')

# Assuming your data is called "d1", a dataframe with genes as rows (labeled as gene symbols),
# and cells are the columns labeled however you like
# Data should be normalized but not scaled
ccAF_HEK293T = ClassifyCells(ccAF, new.data = as.matrix(HEK293T@assays$RNA@data))
names(ccAF_HEK293T) = colnames(HEK293T@assays$RNA@data)
table(ccAF_HEK293T)

Note that Seurat is not required for the classification, only to load up the HEK293T data. If you have an alternate data analysis approach and can get a gene expression matrix to input with gene symbols it should work just as well.