Compute regression-based gene scores

For each gene, regresses its expression on the cell score and uses the regression coefficient (beta) as the gene weight. This evaluates each gene independently and avoids collinearity issues present in the PCA back-projection approach used by computeGeneAndCellScores.

Usage

computeRegressionGeneScores(object, sigma = NULL, verbose = TRUE)

# S4 method for class 'CoPro'
computeRegressionGeneScores(object, sigma = NULL, verbose = TRUE)

# S4 method for class 'CoProMulti'
computeRegressionGeneScores(object, sigma = NULL, verbose = TRUE)

Arguments

object

A CoPro object with cell scores already computed via computeGeneAndCellScores.

sigma

Optional numeric vector of sigma values to process. If NULL (default), all sigma values in object@sigmaValues are used.

verbose

Logical; print progress messages. Default TRUE.

Value

The input object with the @geneScoresRegression slot populated with regression-based gene weights (beta coefficients). The format mirrors @geneScores: a flat list keyed by "geneScores|sigma<value>|<cellType>", each entry a genes x nCC matrix.

Details

Results are stored in the @geneScoresRegression slot, leaving the original PCA-based @geneScores slot untouched.

For each cell type, sigma value, and canonical component (CC):

1. Retrieves the cell score vector from @cellScores.

2. Subsets @normalizedDataSub to cells of that type.

3. Computes beta_g = cov(gene_g, cellScore) / var(cellScore) for every gene g (equivalent to simple linear regression).

4. Stores the beta vector in @geneScoresRegression.

See also

computeGeneAndCellScores for the PCA-based gene scores, testGeneGLM for statistical testing with covariates.