tgvis: Function to estimate and select the optimal number of single effects using profile-likelihood and BIC

This function estimates the number of single effects in a locus by combining profile-likelihood methods and Bayesian Information Criterion (BIC) to optimize the model. It includes resampling for estimating standard errors and performs score tests for infinitesimal effects.

Usage

tgvis(
  estimate_inf = F,
  by,
  bXest,
  LD,
  Noutcome,
  L_vec = c(1:8),
  var_inf = 1e-07,
  estimate_residual_variance = F,
  scaled_prior_variance = 0.5,
  residual_variance = 1,
  max_iter = 50,
  max_eps = 0.001,
  susie_iter = 500,
  pip_thres_cred = 0.95,
  eigen_thres = 0.999,
  varinf_upper_boundary = 0.25,
  varinf_lower_boundary = 0.001,
  ebic_beta = 1,
  ebic_upsilon = 1,
  pip_min = 0.05,
  pv_thres = 0.05,
  pleiotropy_rm = NULL,
  prior_weight_theta = NULL,
  prior_weight_gamma = NULL,
  standization = T
)

Arguments

estimate_inf: An indicator of whether estimating the infinitesimal effect. Default is F.
by: A vector of Z-scores of the marginal effects from the outcome GWAS.
bXest: A matrix of direct effect estimates based on the Z-scores of tissue-gene pairs.
LD: The LD matrix of variants.
Noutcome: The sample size of the outcome GWAS.
L_vec: A vector of candidate numbers of single effects used in BIC. Default is `c(1:8)`.
var_inf: When estimate_inf = F, the variance of infinitesimal effect (estimated by LDSC possibly). Default is 1e-7.
estimate_residual_variance: An indicator of whether of not estimating the variance of residuals in SuSiE. Default is F.
scaled_prior_variance: The prior variance of signals in SuSiE. Default is 0.5 which is slightly larger than 0.2 in SuSiE software.
residual_variance: The residual variance. Default is 1.
max_iter: The maximum number of iterations for the profile-likelihood algorithm. Default is 50.
max_eps: The convergence tolerance for the profile-likelihood algorithm. Default is 1e-3.
susie_iter: The maximum number of iterations for `susie_rss` within the profile-likelihood algorithm. Default is 50.
pip_thres_cred: The cumulative PIP threshold for variables in a credible set. Default is 0.95.
eigen_thres: The threshold of eigenvalues for modelling the infinitesimal effect. Default is 1.
varinf_upper_boundary: The upper boundary for the prior variance of infinitesimal effects, multiplied by var(y) to adapt to different locus variances. Default is 0.25.
varinf_lower_boundary: The lower boundary for the prior variance of infinitesimal effects, not multiplied by var(y). Default is 0.001.
ebic_beta: The extended BIC factor for causal effects of tissue-gene pairs and direct causal variants used in BIC computation. Default is 1.
ebic_upsilon: The extended BIC factor for infinitesimal effects used in BIC computation. Default is 1.
pip_min: The minimum PIP threshold for individual causal effects in the profile-likelihood. This is used to specify which tissue-gene pairs and direct causal variants to include in the score test of variance of infinitesimal effects. Default is 0.05.
pv_thres: The p-value threshold for the score test. Default is 0.05.
pleiotropy_rm: A vector of indices specifying which variants should not be considered as having direct causal effects.
prior_weight_theta: A vector of prior weights of gene-tissue pairs, which will be used as input in SuSiE. Default is NULL.
prior_weight_gamma: A vector of prior weights of direct causal variants, which will be used as input in SuSiE. Default is NULL.
standization: A indicator of whether standardizing the input when performing SuSiE for fine-mapping causal gene-tissue pairs and direct causal variants. Default is T.

Value

A list containing:

theta: The estimated effects for tissue-gene pairs, scaled by the outcome GWAS sample size.
gamma: The estimated effects for direct causal variants, scaled by the outcome GWAS sample size.
theta.pip: Posterior inclusion probabilities (PIP) for tissue-gene pairs.
gamma.pip: Posterior inclusion probabilities (PIP) for direct causal variants.
theta.pratt: Pratt estimations for tissue-gene pairs.
gamma.pratt: Pratt estimations for direct causal variants.
theta.cs: Credible set indicators for tissue-gene pairs.
gamma.cs: Credible set indicators for direct causal variants.
theta.cs.pip: PIP within credible sets for tissue-gene pairs.
gamma.cs.pip: PIP within credible sets for direct causal variants.
upsilon: The estimated infinitesimal effects.
var.upsilon: The estimated variance of infinitesimal effects.
fit.causal: The SuSiE fit object for the causal analysis.
cs.summary: A summary of the credible sets obtained from the analysis.
Bicvec: A vector of BIC values for each candidate number of single effects.