Contents

compute$genome$concordance

Cross-trait directional concordance along the genome

Reduces a PGS library to a per-variant PolyGeniusGenomeSignal measuring whether the models that carry a variant agree on its harmonized effect direction, each model's vote weighted by the magnitude of the weight it places there. The signal stores the signed, sum-reducible pair value.num = sum(oriented beta) and value.den = sum(|beta|), whose ratio S = value.num / value.den lies in [-1, 1].

Usage

compute$genome$concordance(
  models,
  anchor = NULL,
  gwas = NULL,
  reference = NULL,
  outcome = NULL,
  min.models = 2
)

Arguments

Argument	Description
`models`	A PGSLibrary or a single PGS. Anything else aborts, as does an empty PGS library.
`anchor`	One of "outcome", "reference", "none", or NULL (default). NULL auto-selects "outcome" when gwas is supplied, else "reference" when reference is, else "none".
`gwas`	A single-variant [PolyGeniusAssociation](/reference/polygeniusassociation/) (associate$singleVariant() or its meta), or NULL (default). Supplies the per-variant outcome effect that orients + to the risk allele. Required when anchor = "outcome", and aborts when none of its variants overlap models.
`reference`	Character scalar naming one model within models, or NULL (default). Required when anchor = "reference"; aborts when the name is not in models or shares no variant with the other models.
`outcome`	Character scalar, or NULL (default). Selects one outcome when gwas holds several, and is required in that case.
`min.models`	Numeric scalar, default 2. Minimum number of models a variant must appear in to be kept; a variant in one model agrees with itself trivially. Must be at least 1, and the call aborts when no variant clears it with non-zero weight.

Value

A PolyGeniusGenomeSignal with statistic = "concordance", resolution = "variant" and bin.reduce = "sum". $results has one row per retained variant, with chr, position, nea, ea, value.num, value.den and n.models. $metadata$anchor records the sign reference, and $metadata$frame is "relative" under anchor = "reference" and "absolute" otherwise. $diagnostics holds n.ambiguous (strand-ambiguous rows the harmonizer dropped, counted once per model occurrence, including the excluded reference model's own), anchor and reference.

Details

|S| -> 1 when the carrying models agree on direction and that agreement carries real weight; S is near 0 when they are split or the weights are near-null. Because a vote is weighted by |beta|, a shrunken ~0 weight votes ~0, so a split locus reflects genuine cross-trait disagreement rather than the sign of numerical noise. Numerator and denominator are summed and divided per display bin, so the ratio re-aggregates correctly at any bin width.

Low |S| is hypothesis-generating for antagonistic pleiotropy, but it also arises from ordinary cross-trait direction differences in a multi-trait library. Read split loci as candidates, especially where the carrying weight is high.

Anchor -- what the sign means

anchor sets the reference the +/- sign is read against. |S|, the amount of agreement, is the same under every anchor; only the sign's orientation changes.

"outcome" orients + to the outcome risk-increasing allele, taken from gwas. S > 0 means the carrying models agree on pushing outcome risk up, S < 0 protective, near 0 split. This is the reading that ties the track to the outcome the PRS-WAS was run against.
"reference" orients + to the effect direction of the one index model named in reference, which is itself excluded from the vote; S then reads as agreement with that anchor (+1 agrees, -1 opposes).
"none" orients on the canonical (lexicographically larger) allele, so the sign is an arbitrary allele convention with no clinical meaning -- read |S| only.

Examples

# Agreement on the outcome-risk direction, anchored on a single-variant scan
gwas <- associate$singleVariant(data, phenotypes = AD)
sig <- compute$genome$concordance(data$library, gwas = gwas)
visualize$genome$concordance(sig)

# Agreement with one index model instead; the anchor is inferred from `reference`
sig.ref <- compute$genome$concordance(data$library, reference = "LDpred2_AD")