soft_cluster_gam_fit
soft_cluster_gam_fit.RdFit Soft Cluster-Specific GAMs with Ridge Regularization
Arguments
- G
A matrix of gene expression values (genes × cells).
- t
A numeric vector of pseudotime values (length = number of cells).
- P
A soft cluster assignment matrix (cells × clusters), where rows sum to 1.
- df
Degrees of freedom for spline basis (default: 5).
- lambda
Ridge regularization strength (default: 1e-4).
- test
Statistical test to use for model comparison: `"F"` (default) or `"LRT"` (likelihood ratio test).
Value
A list with components:
- stat
A data frame with test statistic, p-value, mean expression difference, and Cohen's d for each gene
- df
Degrees of freedom used for test
- coef
Fitted coefficients from the full model
- design_null
Design matrix for the null model
- design_full
Design matrix for the full model
Details
Fits a generalized additive model (GAM) to gene expression along pseudotime using soft cluster assignments. The model compares a shared smooth (null model) to cluster-specific smooths (full model), using ridge regression and F or likelihood ratio tests.
Examples
# Example pseudotime, profiles, and expression matrix setup
set.seed(123)
t <- seq(0, 1, length.out = 100)
P <- matrix(runif(300), nrow = 100)
P <- P / rowSums(P)
G <- matrix(rnorm(500), nrow = 5)
fit <- soft_cluster_gam_fit(G, t, P, df = 3)
head(fit$stat)
#> stat pval mean_diff cohen
#> 1 0.7847855 0.6172285 0.785680855 0.708225587
#> 2 0.4980950 0.8545586 0.661185867 0.778625761
#> 3 0.7429910 0.6533914 -0.621810882 -0.700928364
#> 4 1.3263654 0.2410829 0.005949389 0.005999336
#> 5 1.5342015 0.1569317 -0.218150387 -0.235564932