Implements the MRLocus colocalization step, in which summary
statistics from A and B studies are used in a generative
model, with a horseshoe prior on the latent true effect sizes.
Posterior effect sizes and posterior SD are returned,
although in pratice we use the original SE in the subsequent
fitSlope
model.
See Supplementary Methods of the MRLocus manuscript.
fitBetaColoc(beta_hat_a, beta_hat_b, se_a, se_b, Sigma_a, Sigma_b, ...)
vector of estimated coefficients for A
" " for B
vector of standard errors for beta_hat_a
" " for beta_hat_b
correlation matrix of SNPs for A
" " for B (this could be different for different LD matrix)
further arguments passed to rstan::sampling
a list with the following elements:
stanfit
object
posterior means for estimated coefficients for A and B
posterior standard deviations for A and B
scaling factors for A and B
Two important notes: (1) in MRLocus manuscript, original
SE are used instead of posterior SD in the slope fitting step,
(2) the posterior means and SD for estimated coefficients
are appropriately scaled, while the results from the
stanfit
object are not scaled. In order to scale the
results from the stanfit object, scale_a
and scale_b
should be divided out from both coefficients and SDs (see Supplementary Methods).
Anqi Zhu*, Nana Matoba*, Emma P. Wilson, Amanda L. Tapia, Yun Li, Joseph G. Ibrahim, Jason L. Stein, Michael I. Love. MRLocus: identifying causal genes mediating a trait through Bayesian estimation of allelic heterogeneity. (2021) PLOS Genetics https://doi.org/10.1371/journal.pgen.1009455