Parallelized Stable Distillation for the OLS Model — distill_pivot

Distill multiple outcome vectors under the same null and alternative OLS model (in parallel)

Usage

distill_pivot_par(y, x, Q, max_num_causal, p_order = seq_len(ncol(x)), ...)

y: a n x m matrix of phenotype vectors, one phenotype per column
x: a n x p matrix of predictors (may be passed as a sparseMatrix, see the Matrix package)
Q: an orthogonal matrix whose columns span the column space of the background covariates
max_num_causal: a non-negative integer, maximum number of causal/active predictors to search for
p_order: a vector of non-negative integers, if provided, columns of x will be distilled in the order p_order[1],p_order[2],..

a list containing

p_value a vector of m p-values from the Renyi Outlier Test, in the same order as the phenotype vectors provided
u a p x m matrix of extracted p-values
signs a p x m matrix of effect signs corresponding the extracted p-values u. Note, these signs are useful for post-hoc interpretation but are NOT extracted in the SD procedure so they are NOT independent of u.
y a n x m matrix of distilled phenotype vectors y left over as a biproduct of distillation (independent of u) and ready for downstream testing

Parallelized implementation of the OLS distillation routine with simple quantile filter proposed in Section 4 of https://arxiv.org/pdf/2212.12539 .