GWAS reproduction tasks

[eric-czech]

This is an umbrella issue to track many of the outstanding tasks for the reproduction:

• BGEN conversion
  • Store post-QC bgen zarr archives for analysis #9
  • Explore using a larger compression block size #7
  • Quantify precision loss in bgen encoding #11
  • Decide on an appropriate chunk size for zarr conversions #6
• Get a Dask cluster working
  • Configure autoscaling for GKE clusters #13
  • Determine how to set resources for Dask workers on GKE #12
  • Figure out how to identify network bottlenecks #18
  • Explore Dask Cloud Provider for deployment #19
  • Getting a GKE cluster working without loadBalancer serviceType #21
  • Upgrade to Zarr 2.5.0 to get async loader Upgrade to zarr 2.5.0 #23
  • Move to VPC Create VPC #33
  • Build GCE VM images Create VM image for Dask CP using Packer  #24
• Address more of the differences in the GWAS script GWAS script tasks #14
• Add betas for comparison in addition to p-values
• Download and prepare summary stats Integrate the OT Neale Lab summary stats #15
• Re-run phenotype prep pipeline (PHESANT)
  • Filter to samples that pass GWAS QC
  • Run all phenotypes instead of subset
• Compare sample sizes (i.e. num complete cases) as well as betas/p-values in merged results
• Expand scope beyond chromosome 21/22:
  • Run rechunking
  • Run QC step 1
  • Run QC step 2
  • Run GWAS
• Clean up and improve documentation Documentation tasks #17
• Determine quantitative measures for validation
  • p-value scatterplots are enough to see clear changes now but at some point more sensitive measures will be useful

Note: https://github.com/pystatgen/sgkit/issues/67 is related but this issue will track tasks less specific to sgkit.