Skip to content

Latest commit

 

History

History
51 lines (37 loc) · 3.5 KB

NEWS.md

File metadata and controls

51 lines (37 loc) · 3.5 KB

vicar 0.1.6

This version mostly changes the documentation. Additions include:

  • Many examples in the main functions in vicar. Including examples in vruv4, ruv3, ruvb, mouthwash, and backwash.
  • Three new vignettes. One providing instructions and examples on how to customize your factor analysis, one providing instructions and examples on how to customize your prior specification in ruvb, and one giving a sample analysis using the functions in vicar and other confounder adjustment packages. Type vignette(package = "vicar") for the list of available vignettes.
  • New functions fa_tester and fa_tester_ruvb for testing whether a user-specified factor analysis is appropriate for the functions in vicar.
  • An example simulated dataset, "sim_gtex", based on the characteristics of the GTEx data. This is so that you can test our your confounder adjustment methods on some good test data. Type data(sim_gtex) to access the data or ?sim_gtex for seeing details about the data.
  • I've removed vruv2 from being exported as it was supplanted by ruv3.

vicar 0.1.5

This version added the function backwash. This is very similar in spirit to mouthwash, except that rather than estimate the confounders by maximum likelihood, backwash does so using a more Bayesian approach. backwash returns a variational approximation to the posterior.

vicar 0.1.4

This version added the function mouthwash to adjust for hidden confounding when one does not have control genes. It applies the same prior from ashr to a factor-augmented regression framework.

vicar 0.1.3

A lot of changes have occurred since my last news update. The biggest changes are:

  • RUV3 is a method that can be considered both a version of RUV2 and RUV4. I implemented this in the function ruv3.
  • ruvimpute is a generic function for using matrix imputation for confounder adjustment.
  • ruvb is a special Bayesian version of RUV. It is highly customize-able, as you can tweak the Bayesian factor analysis and the prior specifications of the effects.
  • I no longer recommend vruv2 as this is now subsumed by ruv3. I'll probably remove vruv2 in the future.

I provide reasonable defaults for all new methods.

vicar 0.1.2

  • vruv2 now works pretty well and is recommended for use. This is a very different way to do variance inflation in RUV2 than what was previously implemented.
  • The previous implementation is now in the function vruv2_old, but it may be removed at any time.
  • I included ash_ruv2 that is a wrapper for vruv2 and ashr::ash.workhorse.
  • Some new factor analyses are available under the hood, but none of them are recommended for general use: pca_ruv2, qmle_ruv2, and pca_2step. In the future, I plan on only saving pca_2step.

vicar 0.1.1

  • Added vruv2, a variance-inflated version of RUV2, but it doesn't work too well yet.
  • The main function for variance-inflated RUV4 is now vruv4. I thought that vicarius_ruv4 was too verbose. In the future, as I create new calibrated versions of confounder adjustment methods, the function name will just have a "v" in front of the name of the confounder adjustment method.
  • To get the standard errors of betahat, vruv4 now multiplies the estimated variances by ([X, Z]'[X, Z])^{-1} rather than (X'X)^{-1}.
  • I plan on adding a separate function for variance inflation without confounder adjustment and removing future capabilities of using vruv4 when k = 0. Keep this in mind.
  • limmashrink = TRUE is now the default.