Notes on Mac OS

For some unknown reasons, compiled Eigen library does not work as expected with the clang compiler. The matrix operations keep producing lots of nan results. We recommend using gcc after version 5 that fully support std=c++14 flag.

Installation (a simple version)

Just do this in R

> install.packages('zqtl_x.x.x.tar.gz')

Installation (optional)

Make sure your R development environment support C++14 by including -std=c++14 to CFLAGS and CXXFLAGS in ~/.R/Makevars file. For instance,

CXX = g++-6
CXXFLAGS = -O3 -std=c++14
CFLAGS = -O3 -std=c++14

Build package locally.

$ R CMD build .

You will have zqtl_x.x.x.tar.gz gzipped file in current directory (x.x.x can be any version of the package). Install package within R:

> install.packages('zqtl_x.x.x.tar.gz')

To speed up matrix-vector multiplication, one can compile codes with Intel MKL library. Currently RcppEigen does not support BLAS only options (this may not be true).

Enjoy!

Release notes

v 1.4.2

  • Add faster SVD routine for a large genotype matrix
  • Bugfix in randomized SVD

v 1.4.1

  • Regularization of eigen values (the eigen.reg parameter)

v 1.4.0

  • Add useful utilities

v.1.3.6

  • Improve variance calculation
  • Partitioned variance calculation with annotation information

v.1.3.5

  • Report both types of residual variance estimation
  • Lower the precision limit in SGVB steps
  • Add adjusting routines for convenience

v.1.3.4

  • Remove “backfire control” and “two step optimization”
  • Revive variance calculation with residual estimation
  • Make a room for univariate confounding factors
  • Output “clean” version of GWAS effects

v.1.3.3

  • Minor fix on smoothness of the unmediated effect
  • Add do.control.backfire to regress out M0 -> M.
  • Add experimental factorization model factorization.model = 1

v.1.3.2

  • Estimate unmediated factors by factorization
  • Boost sample size by resampling eigen vectors
  • Allow two step optimization
  • Parametric bootstrap for sensitivity analysis

v.1.3.1

  • Make propensity sampling as second option
  • Estimate the unmediated effect as before
  • Drop weight features (not so useful)
  • Allow multiple mediators in the conditional analysis

v.1.3.0

  • Counterfactual estimation of average unmediated effects via sampling
  • Factored mediation model

v.1.2.3

  • Take multivariate effect sizes for mediator QTLs
  • Initialization by dot product

v.1.2.2

  • Restrict number of mediator variables during the estimation of unmediated effects

v.1.2.1

  • Estimation of average unmediated effects
  • Minor bugfix in regression (twice rescaling)

v.1.2.0

  • Matrix factorization for confounder correction
  • Confirmed usefulness of non-negative parameters
  • Simplified pleiotropy model in mediation analysis
  • Adjust scales by standard deviation
  • Variance model in mediation analysis

v.1.1.0

  • Spiked Gamma for the factored zQTL methods
  • Additional covariate component for confounder correction

v.1.0.1

  • Variance calculation. See note/variance.md for details.
  • Random effect component to account for uncertainty of individuals
  • Removing LD-structure bias between mediation QTL and GWAS cohorts

v.1.0.0

  • Initial version migrated from MIT github

Bug reports

Yongjin Park ypp@stat.ubc.ca