Contents

visualize$associations$landscape

Association landscape across trait groups

One point per PRS-level association, grouped along the x-axis by group.by and placed at the -\log_{10} of the p-value column pval names. The PGS-wide association scan's counterpart to a Manhattan plot, with trait groups in place of chromosomes.

Usage

visualize.associations.landscape(
  associations,
  group.by,
  pval = adj.pval,
  statistic = c("neglog10p", "signed.neglog10p"),
  threshold = 0.05,
  label.top = NULL,
  outcome = NULL,
  palette = NULL,
  trans = NULL,
  theme = c("polygenius", "none")
)

Arguments

Argument	Description
`associations`	A PolyGeniusAssociation of PRS-level rows, from [associate$regression](/reference/associate-regression/) or a [associate$meta](/reference/associate-meta/) over one. A schema that does not declare the "landscape" plot aborts, as does a meta whose source.schema does not declare it (a single-variant or mediation meta).
`group.by`	Unquoted expression giving each row's group, e.g. group.of[predictor]. Must resolve to one value per row of $results. Group order follows its factor levels when it is a factor, otherwise first appearance.
`pval`	Unquoted expression for the p-value column, default adj.pval.
`statistic`	One of "neglog10p" (default), "signed.neglog10p". Height: -\log_{10}(p), or that value signed by estimate so risk and protective associations separate above and below zero.
`threshold`	Numeric scalar, default 0.05, or NULL for no line. A dashed line at -\log_{10}(threshold), mirrored below zero when signed.
`label.top`	Integer scalar, or NULL (default). Label the rows with the largest absolute height by predictor, qualified by outcome when more than one outcome is drawn.
`outcome`	Character vector, or NULL (default, every outcome). Outcomes to draw, matched against the outcome column.
`palette`	Group colours: a palette-system or hue name, a vector of colours, or NULL (default) for the package categorical palette. Above 30 groups the first two colours of a vector are used as the two tones.
`trans`	One of "sqrt", "log10", or NULL (default, a linear axis). Y-axis transform, to spread the rows near the threshold instead of letting a few strong associations flatten them. "log10" is scales::pseudo_log_trans(), symmetric about zero, with ticks at powers of ten. "sqrt" needs non-negative heights, so it refuses a signed statistic.
`theme`	One of "polygenius" (default), "none". Plot theme. "none" gives a bare theme_minimal() to style yourself; palette colors are applied either way.

Value

A ggplot object.

Details

Only term.type == "main" rows are drawn when the table carries that column. group.by and pval are evaluated against $results before any row is dropped, so an expression such as group.of[predictor] may use a lookup vector from the calling environment.

The plot reads pval as given. It never adjusts a p-value, so the threshold line is only as meaningful as the column it is drawn against: with the default adj.pval, threshold = 0.05 is an FDR cut. Rows whose height is not finite are dropped; a p-value that underflowed to zero is dropped with a warning, and the rendered subtitle gives the count. Points are jittered within their group by a seeded position_jitter(), so repeated renders place them identically.

Up to 30 groups are coloured from the categorical palette. Beyond that, groups alternate between two tones, as chromosomes do on a Manhattan track.

Examples

assoc <- structure(list(results = data.frame(
  schema = "glm", predictor = c("PRS.AD", "PRS.T2D", "PRS.LDL"),
  outcome = "dementia", term.type = "main", estimate = c(0.4, 0.05, -0.1),
  pval = c(1e-8, 0.3, 0.01), adj.pval = c(3e-8, 0.3, 0.015))),
  class = c("PolyGeniusAssociation", "list"))
group.of <- c(PRS.AD = "neuro", PRS.T2D = "metabolic", PRS.LDL = "metabolic")
visualize$associations$landscape(assoc, group.by = group.of[predictor],
                                 statistic = "signed.neglog10p")