Currently, when we match LD matrices with GWAS summary statistics, a new matrix is created in temp_dir with only the subset of SNPs that are shared between the two data sources. This can be very slow and expensive for large LD matrices, potentially replicating 100s of GB of data unnecessarily.
As an alternative solution, we can instead generate a mask that ensures that the posterior estimates for SNPs that only exist in the LD matrix is set to zero by default. This mask would be generated in the matching step and would be passed down for downstream tasks, such as PRS model fitting.
Currently, when we match LD matrices with GWAS summary statistics, a new matrix is created in
temp_dir
with only the subset of SNPs that are shared between the two data sources. This can be very slow and expensive for large LD matrices, potentially replicating 100s of GB of data unnecessarily.As an alternative solution, we can instead generate a mask that ensures that the posterior estimates for SNPs that only exist in the LD matrix is set to zero by default. This mask would be generated in the matching step and would be passed down for downstream tasks, such as PRS model fitting.
Details TBD.