PSPS experiments

Simulations

All simulations are done using R with version 4.2.1 (2022-06-23). The required packages are POPInf, IPD, quantreg, doParallel, data.table, knockoff, AER, MASS, sandwich, and lmtest.

Tasks that have been implemented for ML-assisted inference

• ./old_task/Fun.R: Functions for simulation and PSPS protocol
• ./old_task/mean.R: mean estimation
• ./old_task/ols.R: linear regression
• ./old_task/logistic.R: logistic regression

Tasks that have not been implemented for ML-assisted inference

• ./new_task/Fun.R: Functions for simulation and PSPS protocol
• ./new_task/qr.R: quantile regression
• ./new_task/ngbr.R: negative binomial regression
• ./new_task/iv.R: instrumental variables regression
• ./new_task/dlasso.R: debiased Lasso
• ./new_task/ranksum_t1e.R: Wilcoxon rank-sum test (type-I error)
• ./new_task/ranksum_power.R: Wilcoxon rank-sum test (power)

FDR control

• ./fdr/Fun.R: Functions for simulation and PSPS protocol
• ./fdr/Fun_PSPS-knockoff.R: Functions for PSPS-knockoff
• ./fdr/BH: PSPS-BH simulation
• ./fdr/data_knockoff: generate data for PSPS-knockoff simulations
• ./fdr/dlasso_knockoff: fit PSPS-dlasso on the generated data
• ./fdr/summarize_knockoff: fit PSPS-knockoff

Real data analysis

The real data analysis is done in UK Biobank, which is avaiable by application this link. We used plink2 and QUAIL for fitting quantile regression to identify vQTL.

• ./real/Fun.R: Functions for PSPS protocol
• ./real/vQTL.sh: run genome-wide vQTL analysis
• ./real/combine.R: apply PSPS protocol