scCAN {scCAN}

scCAN

Description

This is the main function to perform sc-RNA seq data clustering clustering. scCAN is fully unsupervised scRNA-seq clustering framework that uses deep neural network and network fusion-based clustering algorithm. First, scCAN applies a non-negative autoencoder to filter scRNA-seq data. Second, the filtered data is passed to stacked Bayesian autoencoder to get multiple low-dimensional representations of input data. Subsequently, scCAN converts these compressed data into networks and unify those networks to a single graph. Then, scCAN uses a spectral clustering algorithm to obtain final clusters assignment.

Usage

scCAN(
  data,
  sparse = FALSE,
  n.neighbors = 30,
  alpha = 0.5,
  n.iters = 10,
  ncores = 10,
  r.seed = 1,
  subsamp = TRUE,
  k = 2:15,
  samp.size = 5000
)

Arguments

data

Gene expression matrix, with rows represent samples and columns represent genes.

sparse

Boolen variable indicating whether data is a sparse matrix. The input must be a non negative sparse matrix.

n.neighbors

Number of neighboring cells that are used to caculate the edge's weight. The number of neighbors are set n.neighbors = 30 by default.

alpha

A hyper parameter that control the weight of graph. This values is set to alpha = 0.5 by default.

n.iters

A hyper-parameter to set the number of network fusion iterations. It is set to n.iters = 10 by default.

ncores

Number of processor cores to use.

r.seed

A parameter to set a seed for reproducibility. This values is set to r.seed = 1 by default.

subsamp

Enable subsampling process for big data. This values is set to subsamp = T by default.

k

A vector to search for optimal number of cluster.

samp.size

A parameter to control number of sub-sampled cells.

Value

List with the following keys:

• cluster - A numeric vector containing cluster assignment for each sample.

• k - The optimal number of cluster.

• latent - The latent data generated from autoencoders.

References

1. Duc Tran, Hung Nguyen, Bang Tran, Carlo La Vecchia, Hung N. Luu, Tin Nguyen (2021). Fast and precise single-cell data analysis using a hierarchical autoencoder. Nature Communications, 12, 1029. doi: 10.1038/s41467-021-21312-2

Examples

## Not run: 
# Not run if scDHA has not installed yet.
# Load the package and the example data (SCE dataset)
library(scCAN)
#Load example data
data("SCE")

#Get data matrix and label
data <- t(SCE$data); label <- as.character(SCE$cell_type1)

#Generate clustering result, the input matrix has rows as samples and columns as genes
result <- scCAN(data, r.seed = 1)

#Get the clustering result
cluster <- result$cluster

#Calculate adjusted Rand Index
ari <- round(scCAN::adjustedRandIndex(cluster,label), 2)
message(paste0("ARI = ", ari))


## End(Not run)