Replication of an mQTL analysis using the LOCUS method on simulated data

Overview

A molecular quantitative trait locus analysis is performed using our variational inference procedure for combined predictor and outcome selection implemented in the R package locus. This example reproduces the application given in H. Ruffieux, A. C. Davison, J. Hager, I. Irincheeva, Efficient inference for genetic association studies with multiple outcomes, Biostatistics, 2017.

Data

As the genotying datasets cannot be provided for privacy reasons, we simualted SNPs and outcomes using the R package echoseq to try to emulate the real (lipid)-mQTL data analyzed. The simulated dataset can be downloaded here.

Algorithm

The package locus used for the analysis may be installed with the devtools command

devtools::install_github("hruffieux/locus", ref = "v0.5.0")

where ref = v0.5.0 indicates the git tag corresponding to the package version we used.

IMPORTANT NOTE: this is an old version of the package, which corresponds to that used in the above-cited paper. The newest version of the package has improved scalability and memory management and should be used instead for new analyses.

Workflow

The scripts should be executed in the following order:

1. replicate_real_data_pb.R (analysis using locus) and replicate_real_data_pb_varbvs.R (analysis applying the single-trait variational method "varbvs" by Carbonetto and Stephens, 2012, Bayesian Analysis 7);

2. replicate_perm.R (analysis on permuted data with locus) and replicate_perm_varbvs.R (idem but with varbvs); and

3. replicate_FDR_estimation.R (FDR-based comparison of locus and varbvs) and manhattan.R (manhattan plots for inference with locus and varbvs).

Issues

To report an issue, please use the issue tracker at github.com.