Single-effect regression from summary statistics

Fit a single-effect regression (SER) model directly from marginal summary statistics. Unlike susie_rss, this interface does not take an LD matrix or reference factor matrix, and it never constructs a diagonal LD matrix internally. The model has exactly one single effect.

Usage

susie_ser(
  z = NULL,
  bhat = NULL,
  shat = NULL,
  n = NULL,
  var_y = NULL,
  prior_variance = 50,
  scaled_prior_variance = 0.2,
  prior_weights = NULL,
  null_weight = 0,
  estimate_prior_method = c("optim", "simple"),
  check_null_threshold = 0,
  prior_tol = 1e-09,
  coverage = 0.95,
  ethres = NULL
)

Arguments

z: A vector of z-scores. Provide either z, or both bhat and shat, but not both.
bhat: Alternative summary statistics: marginal effect estimates.
shat: Standard errors for bhat; may be a scalar or a vector with the same length as bhat.
n: Optional sample size. When supplied, z-scores are adjusted using the same PVE adjustment as susie_rss.
var_y: Optional sample variance of the outcome. When supplied with n, this sets the working outcome scale. With bhat and shat, coefficients returned by coef.susie are on the original bhat scale.
prior_variance: Prior variance on the z-score scale when n is not supplied. Default 50, matching the z-only RSS scale.
scaled_prior_variance: Prior variance divided by the working outcome variance when n is supplied. Default 0.2.
prior_weights: A vector of prior probabilities, one per variable. The weights are normalized internally.
null_weight: Prior probability of no effect. If nonzero, a null variable is appended with zero effect.
estimate_prior_method: Method used for the single prior variance. "optim" optimizes the one-effect marginal likelihood; "simple" uses the supplied prior variance and applies the null-threshold check.
check_null_threshold: If the log-likelihood at prior variance zero is within this amount of the selected prior variance, set the prior variance to zero.
prior_tol: Prior-variance tolerance used by susie_get_pip.
coverage: Coverage for credible-set construction. If NULL, credible sets are not constructed and no no-LD hint is emitted.
ethres: Effective-number threshold passed to susie_get_cs_attainable.

Value

A "susie" object with one single effect.

Single-effect regression from summary statistics

Usage

Arguments

Value

See also