Multivariate Bayesian variable selection regression

Further investigation of mismatched analysis with identity prior

I observe unexpected FDR inflation for analyzing singleton simulations using identity priors. Here I take a closer look at the problem.

In [1]:
%cd ~/GIT/github/mnm-twas/dsc/finemap_output

/scratch/midway2/gaow/GIT/github/mnm-twas/dsc/finemap_output
In [16]:
scenario = 'singleton/small_data_1_singleton_1.pkl'
analysis = 'mnm_identity/small_data_1_singleton_1_oracle_generator_1_mnm_identity_1.rds'
In [17]:
dat = dscrutils::read_dsc(scenario)
res = readRDS(analysis)$result
In [18]:
names(dat)
  1. 'Y'
  2. 'R'
  3. 'J'
  4. 'meta'
  5. 'DSC_DEBUG'
In [19]:
names(res)
  1. 'alpha'
  2. 'mu'
  3. 'mu2'
  4. 'KL'
  5. 'lbf'
  6. 'sigma2'
  7. 'V'
  8. 'elbo'
  9. 'niter'
  10. 'fitted'
  11. 'coef'
  12. 'null_index'
  13. 'mixture_weights'
  14. 'lfsr'
  15. 'intercept'
  16. 'sets'
  17. 'pip'
  18. 'm_init'
In [20]:
# plot = mvsusieR::mvsusie_plot(res)
In [21]:
truth = dat$meta$true_coef
true_idx = which(truth != 0, arr.ind = TRUE)
true_idx
rowcol
1361
In [22]:
truth[unique(true_idx[1]),]
  1. -0.452527792535091
  2. 0
  3. 0
  4. 0
  5. 0
In [23]:
res$coef[unique(true_idx[1])+1,]
  1. -0.0320290020689475
  2. 0.00611172370821163
  3. 0.000999751801141419
  4. 1.43997799969134e-05
  5. 0.00491099617805023
In [24]:
res$sets
$cs
$L1 =
  1. 63
  2. 119
  3. 128
  4. 130
  5. 136
$purity
min.abs.corrmean.abs.corrmedian.abs.corr
L10.96850250.98505410.9828454
$cs_index
1
$coverage
0.95
Copyright © 2016-2020 Gao Wang et al at Stephens Lab, University of Chicago