scCoNet: Copulas with Zero-One Beta Mixture Margins for Single-Cell Covariation Networks

scCoNet is an R package for modeling zero-one-inflated proportion data using Frank copula models with mixture margins. It implements a two-stage maximum likelihood estimation (tsMLE) procedure and a two-stage likelihood ratio test (tsLRT) for estimating the copula dependence parameter, which can then be used to build covariation networks.

scCoNet uses a mixture margin copula model for pairs of zero-one-inflated beta distributions for gene data $(X_i, X_j)$:

$$ F(x_i, x_j; \gamma_i, \gamma_j, \theta_{ij}) = C\Big( F_i(x_i; \gamma_i),; F_j(x_j; \gamma_j);; \theta_{ij} \Big) $$

where:

• $F_i$ and $F_j$ represent the marginal distribution functions with mixture margins.
• $\theta_{ij}$ is the copula dependence parameter that captures the covariation between the two features.

This model allows the estimation of dependence parameters that serve as the basis for constructing covariation networks.

To install the latest version of scCoNet from GitHub, use the following command:

install.packages("devtools")
devtools::install_github("Kangyi0406/scCoNet")

Once installed, you can access the documentation for any function via ? in R. For example:

library("scCoNet")
?scCoNet

# Load your data (gene values for each cell or feature)
data(gene_data)
data <- gene_data$gen_data

# Load covariates (e.g., experimental conditions or other metadata)
cov <- gene_data$cov_data
cov$name_data <- factor(cov$name_data)
cov$year <- as.numeric(cov$year)
cov <- cov[, 2:3]  # Select the necessary columns

# Run the scCoNet function
result <- scCoNet(
  data,
  covars = data.frame(cov),
  ncores = 1,
  formula.mu = y ~ name_data + year,
  formula.sigma = y ~ name_data + year,
  formula.nu = y ~ name_data + year,
  sig_level = 0.05,
  sig_connect = TRUE
)

1. data: A $n \times p$ data frame of gene expression data with rows as samples and columns as genes.
2. covars: A $n \times q$ data frame of covariates (e.g., experimental conditions) with rows in the same order as data
3. ncores: The number of cores for parallelization. Default is 1 (no parallel computing).
4. formula.mu: A formula for the mean parameter in the marginal regression model.
5. formula.sigmaA formula for the dispersion parameter in the marginal regression model.
6. formula.nu A formula for the zero-one inflation probability parameter.
7. formula.tau A formula for the zero-one inflation probability parameter.
8. sig_level Significance threshold (after Benjamini–Hochberg adjustment) used to determine which associations are retained. Default is 0.05.
9. sig_connect If TRUE, only statistically significant associations are returned; otherwise, all estimated connections are included. Default is FALSE.

For bug reports, issues, or suggestions, please contact the maintainer: Kangyi Zhao via email.