Question: correct approach for create_mash_prior from data?

[thouis]

When using create_mash_prior with sample_data, what is the correct approach for eQTL data in multiple conditions?

In our case, we have eQTL data in a set of donors across a few (< 10) conditions, which we have used SuSiE to finemap individually. Should sample_data be all the betas & SEs for all SNPs, the lead SNPs for each gene across conditions, the set of fine-mapped across conditions, or something else? We'd like to use mvsusieR to fine-map across the conditions simultaneously.

Thanks for any guidance.

[thouis]

It looks like create_mash_prior has been removed. Is there a new recommended way to create the prior?

[stephens999]

sorry, we are working on better documentation and interface.
This is the most recent vignette:
https://stephenslab.github.io/mvsusieR/articles/mvsusie_intro.html
which uses create_mixture_prior to create the prior.

To address the question you asked originally though: I would suggest using the
lead SNP for each gene across conditions to form your prior.

Please bear with us as we work to improve the documentation.

[thouis]

Thank you. That's enough to get us started. We really appreciate you all making the code available early.

[thouis]

We had some trouble passing the mash prior to create_mixture_prior. Currently, we're doing this:

newUlist <- mashr:::expand_cov(m.c_strong$fitted_g$Ulist, m.c_strong$fitted_g$grid, usepointmass=TRUE) prior_mixt <- create_mixture_prior(list(matrices = newUlist[-1], weights = m.c_strong$fitted_g$pi[-1]), null_weight = m.c_strong$fitted_g$pi[1])

where m.c_strong is the result from mashr::mash. We're not certain this is correct. We based this on how create_mash_prior used to work in the codebase, so it's highly speculative on our part.