Calibrate a cohort-specific dominance threshold — fit_dominance

Calibrates a cohort-specific threshold for ws_dominance_flag as the (1 - alpha) quantile of ws_dominance_score computed on a training reference cohort. Recommended over the universal 0.60 default, which is provided only for backwards compatibility.

Usage

fit_dominance_threshold(x_train, top_k = 2L, alpha = 0.05, pseudocount = 0.5)

Arguments

x_train: Samples x features matrix of the training reference cohort (typically Tier R, control samples only).
top_k: Integer; passed to ws_dominance_score. Default 2.
alpha: Numeric in (0, 1). Threshold is the (1 - alpha) quantile of the training dominance distribution, i.e. samples in the top alpha fraction of the training distribution are flagged. Default 0.05.
pseudocount: Numeric; passed to ws_dominance_score. Default 0.5.

Value

List with components threshold, alpha, top_k, reference_n, reference_quantiles.