maybe we should not output lfdr?

[stephens999]

in results we do output a matrix of lfdr values. These will tend to be sensitive to misestimated covariance matrices, and generally we would not recommend users to use them - they should use the lfsr. The vignettes all encourage use of lfsr, and never mention lfdr but people nonetheless will see the lfdr if

they look at names(results). Maybe we should not even be outputting them, at least by default. (If we need them for testing purposes we could add a flag and tell people not to use it in the documentation...)

[pcarbo]

@stephens999 Shouldn't we be consistent with what ashr does? ashr::ash outputs both lfdr and lfsr [and help(ash) does not provide guidance on this].

[stephens999]

the issue i'm worrying about is that lfdr is even more sensitive to prior for mash than for ash... so in ash lfdr is arguably often ok. for mash i would tend never to use it.

[pcarbo]

@stephens999 I added a new argument, mash(..., output_lfdr = FALSE). This is what help(mash) now says:

output_lfdr: If output_lfdr = TRUE, output local false discovery rate estimates. The lfdr tends to be 
sensitive to mis-estimated covariance matrices, and generally we do not recommend using them; 
we recommend using the local false sign rate (lfsr) instead, which is always returned, even when 
output_lfdr = TRUE.

[stephens999]

looks great, thank you!

On Thu, Apr 8, 2021 at 9:08 AM Peter Carbonetto @.***> wrote:

@stephens999 https://github.com/stephens999 I added a new argument, mash(..., output_lfdr = FALSE). This is what help(mash) now says:

output_lfdr: If output_lfdr = TRUE, output local false discovery rate estimates. The lfdr tends to be sensitive to mis-estimated covariance matrices, and generally we do not recommend using them; we recommend using the local false sign rate (lfsr) instead, which is always returned, even when output_lfdr = TRUE.

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/stephenslab/mashr/issues/93#issuecomment-815855006, or unsubscribe https://github.com/notifications/unsubscribe-auth/AANXRRLORBVTVUV6BJMKYADTHW2EHANCNFSM42RRMC2Q .