Prior subsetting

[Marwansha]

I am running mvSuSiE on eQTL data from 15 conditions, where I have matrix eQTL results per gene per region. I am unsure whether the prior covariance matrices should be estimated once on the whole dataset (all genes/regions combined) or computed per gene/feature using only the summary statistics within that specific region.

Additionally, if I estimate the prior on the whole data but a specific feature/gene is missing 1 or 2 conditions, is it acceptable to subset the precomputed prior to the available conditions for that feature,

Any guidance on what is more statistically recommended would be greatly appreciated.

Marwan

[Marwansha]

or is it more recommended to compute prior per feature?

thanks