compute

Namespaced entry point for every transformation that reads or writes a PolyGeniusStudy's sample, model or score axes. Its components:

• compute$scores(): Calculate PRS scores from genotypes
• compute$scores.plan(): A-priori resource estimate (peak RAM, batch count, strategy) for compute$scores() -- never a score matrix; optionally sweeps a memory budget
• compute$standardize(): Build a per-model centered/scaled score layer, tagged with its (center, scale, scale.source)
• compute$populationStructure(): Compute principal components for population stratification correction
• compute$relatedness$kinship(): Sparse KING kinship over related sample pairs
• compute$relatedness$prune(): Samples to retain after removing relatives
• compute$similarity$samples(): Compute sample-sample similarity matrices
• compute$similarity$models(): Compute model-model similarity matrices
• compute$embedding$samples(): Compute low-dimensional embeddings for samples
• compute$embedding$models(): Compute low-dimensional embeddings for models
• compute$genome$concordance(): Per-variant cross-trait effect-direction concordance (anchored)
• compute$genome$convergence(): Per-variant convergent outcome-signal (Sigma|attribution|)
• compute$genome$cumulativeWeight(): Per-variant cumulative |beta| across the model library
• compute$genome$attribution(): Per-(variant,model) attribution (weight x single-variant effect)

A typical compute workflow:

1. Calculate PRS scores:

compute$scores(data, maf.thr = 0.01)

2. Compute population structure:

compute$populationStructure(data,
                            npcs = 10,
                            reference.panel = "EUR")

3. Resolve relatedness (before PCA -- components are distorted by families):

data$sample.pairs$kinship  <- compute$relatedness$kinship(data, degree = 2)
data$samples$unrelated <- compute$relatedness$prune(data, degree = 2)
data <- data[unrelated, ]

4. Compute similarity and embeddings:

# Sample similarity
sim.samples <- compute$similarity$samples(data, X, method = "pearson")
# Model similarity
sim.models <- compute$similarity$models(data, X, method = "snp.jaccard")
# PCA embedding
pca.emb <- compute$embedding$samples(data, X, method = "pca", n.components = 10)
# UMAP embedding
umap.emb <- compute$embedding$samples(data, X, method = "umap")

Single-variant association (GWAS) lives in associate$singleVariant(), not here.