storeylab / ffdr Goto Github PK

Current vignette is extremely out of date (uses entirely different functions and arguments)

There are still warnings (mostly small stuff) when R CMD TEST is run on the package; needs to be handled before it is merged with qvalue.

The functionality in fFDR is intended to be combined with the qvalue package. Some notes:

• This should be done only when/if qvalue is converted to S4 and unit tests are added
• One package can contain multiple vignettes, which means there will be a separate vignette for the use of fqvalue as opposed to qvalue
• This will likely be done on a branch of qvalue in the private repository

estFPi0 currently returns an "fPi0" S4 object; fqvalue function should do the same.

While my initial approach to monotonic smoothing in 2d was to estimate a square grid and perform isotonic regression on each, I've found this does not play nicely with locfit's 2d density estimation (there are consistent and dangerous artifacts produced). Not enforcing any kind of monotonicity can lead to weird behavior where large p-values are accepted and low ones rejected.

Should include text explaining the caveat with fpi0 that this is the conservative choice (and will still outperform traditional FDR)

This should be last step before passing on to John.

library(fFDR)
ttests_example <- simulate_t_tests(m = 2000)
p.value <- ttests_example$p.value
z <- ttests_example$n
fq <- fqvalue(p.value, z)
p <- plot(fq)
grid::grid.draw(p)

Need to use grid::grid.draw . Not sure if by design. Also, plotting functions are not mentioned in vignette

Some fFDR functions don't have examples, some do for deprecated versions of the function that no longer work. All exported functions need to have at least one line of example code.

Simulations are currently a mix of regular data.table objects and S3 functions. Need to set them up to be S4 compatible; is important both for the package itself and for demonstrations in the manuscript.

Should cover

• multiple arguments, including different methods
• plotting and summary methods for resulting object

Hello,

I am new to R but wanted to try this package. I have received the following error using R studio 12 on the most updated current R version:

Warning messages:
1: package ‘qvalue’ is not available for this version of R

A version of this package for your version of R might be available elsewhere,
see the ideas at
https://cran.r-project.org/doc/manuals/r-patched/R-admin.html#Installing-packages
2: In i.p(...) :
installation of package ‘C:/Users/eliav/AppData/Local/Temp/RtmpSMAaOE/file33083a57e5a/fFDR_0.1.tar.gz’ had non-zero exit status

Any pointer would be greatly appreciated!
Thanks
E

These should examine:

• Multiple # of hypotheses m
• Multiple pi0 values
• Multiple sample standard deviations (affects power)
• Large number of replications (done on cluster), each with same chosen mu

Note that issues with the "manuscript" label serve as a personal to-do list for the manuscript (which is not actually on GitHub), not the software.

Right now it's just in fpi0