scTenifoldNet (version 1.4)

pcNet: Principal Component Network Analysis

Description

Computes a gene co-expression network using Principal Component Regression. Each gene is regressed against principal components derived from all other genes.

Usage

pcNet(
  X,
  nComp = 3,
  scaleScores = TRUE,
  symmetric = FALSE,
  q = 0,
  priorNetwork = NULL,
  verbose = FALSE,
  nCores = 1,
  useRcpp = TRUE
)

Value

A sparse matrix (Matrix::dgCMatrix) representing the gene network. Entry [i,j] contains the regression coefficient from gene i to gene j. Diagonal is always zero (no self-loops).

Arguments

X

A matrix with genes as rows and cells/samples as columns. Can be a regular matrix or dgCMatrix (sparse). Must have positive row sums (quality control applied).

nComp

Number of principal components to use for regression. Must be >= 2 and < number of genes. Default: 3.

scaleScores

If TRUE (default), scales output network by maximum absolute value to normalize edge weights to [-1, 1].

symmetric

If TRUE, symmetrizes the network matrix as (A + t(A))/2. Default: FALSE.

q

Quantile threshold for filtering edges. Values below this quantile are set to zero. Range: [0, 1]. Default: 0 (no filtering).

priorNetwork

A data.frame containing a prior gene regulatory network. The data.frame must have two columns: `regulators` and `targets`. Default: NULL.

verbose

If TRUE, prints progress updates. Default: FALSE.

nCores

Number of cores for parallelization. If > 1, uses future backend (multisession). Default: 1 (sequential).

useRcpp

If TRUE (default), uses compiled Rcpp backend for faster computation. Falls back to pure R if Rcpp backend unavailable. Default: TRUE.

Details

Algorithm: 1. Standardize input matrix (center and scale by columns) 2. For each gene K: a. Extract all other genes as design matrix (Xi) b. Compute truncated SVD of Xi to get nComp principal components c. Project gene K onto these components d. Perform OLS regression to get coefficients 3. Assemble coefficients into sparse network matrix 4. Apply optional scaling and filtering

Examples

if (FALSE) {
  # Create sample data
  X <- matrix(rpois(1000*50, 5), nrow=1000, ncol=50)
  rownames(X) <- paste0('Gene', 1:1000)

  # Compute network
  net <- pcNet(X, nComp=3, nCores=4)
}