interface with mashr

[stephens999]

we should document how to use this with mashr, probably in the mashr vignette?
It may require writing a function or functions to produce mashr objects (priors?) in the udr package?

[pcarbo]

Yes, that's a good idea, but perhaps it makes sense to tackle Issue #21 first, and then illustrate the (more user-friendly) interface in the vignette.

[stephens999]

related to stephenslab/mashr#132

• does udr handle missing data?
• Can we fit both the EE and EZ model with udr?

[yunqiyang0215]

@stephens999 udr handles missing data using mashr function. The input to ud_init() can be a mashr object, and handling missingness part is using mashr.

We can fit both EE and EZ model with udr.

[stephens999]

In mashr, V is the correlation matrix, so the covariance is SVS. Is it the same in udr software? (I'm asking because in the udr paper we use V for the covariance) .

…

On Tue, Nov 5, 2024, 13:41 yunqiyang0215 ***@***.***> wrote: @stephens999 <https://github.com/stephens999> udr handles missing data using mashr function. The input to ud_init() <https://github.com/stephenslab/udr/blob/master/R/ud_init.R> can be a mashr object, and handling missingness part is using mashr. We can fit both EE and EZ model with udr. — Reply to this email directly, view it on GitHub <#22 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AANXRRLLM7L7LQEGJRB77P3Z7EGO3AVCNFSM6AAAAABRA6VHGWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDINJXHEYDSMBVGM> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[yunqiyang0215]

I think V is for covariance for udr.

[stephens999]

i think we should think a bit about this.
If we keep that different convention in the two packages
then we should probably not mix the mash data object with the udr V as inputs to the same function.

For example, currently we allow both a mash_data object and V to be provided to udr_init, but
then it's confusing: not clear if the V refers to the udr V or the mashr V.
Maybe it would be enough to not allow the user to specify both?

Or maybe it would be better to be consistent about the use of V in the two packages?

[yunqiyang0215]

Agree... Would be better to be consistent about the notation V. The only concern is if we change udr software, the notation will be different from what we wrote in paper.

[stephens999]

it turns out the same is true for mashr: we use V for variance in the paper, but for correlation in the software

[stephens999]

@pcarbo do you have thoughts on how to proceed?

[pcarbo]

We should do our best to make sure the udr interface is consistent with the existing mashr interface. (This doesn't necessarily mean we need to change the variable names in the underlying implementation. But the input arguments in the exported functions should be consistent.)

[stephens999]

the mashr interface is built around the notion of a "data" object, which contains Bhat, Shat (optional; default is 1) and a correlation matrix (or list of correlation matrices) V.
Do we want to build udr around the same type of object?

[pcarbo]

Definitely!