GWAS script tasks

[eric-czech]

TODOs in the actual GWAS script:

• Regress on complete cases only
  • This is not explicit in the Neale Lab code but it is implicit in the linreg3 call at https://github.com/Nealelab/UK_Biobank_GWAS/blob/master/imputed-v2-gwas/7_run_linreg3.py#L63 and https://github.com/Nealelab/UK_Biobank_GWAS/blob/master/0.1/22.run_regressions.py#L92
  • From the linreg3 docs:

    linreg3() uses the same set of samples for each phenotype, namely the set of samples for which all phenotypes and covariates are defined.

• Add interaction terms in covariates
  • These were added as they are in https://github.com/Nealelab/UK_Biobank_GWAS/blob/master/0.1/22.run_regressions.py#L71
• Determine how many PCs to use
  • 20 are used in https://github.com/Nealelab/UK_Biobank_GWAS/blob/master/0.1/22.run_regressions.py#L71
  • The Details and Considerations post (which pertains to the earlier results from the code in imputed-v2-gwas) states:

  To simplify the process of association testing, association for all phenotypes used a least-squares linear model predicting the phenotype with an additive genotype coding (0, 1, or 2 copies of the minor allele), with sex and the first 10 principal components from the UK Biobank sample QC file as covariates.

  • Conclusion: 20 PCs will be used
• Determine which sex + age fields we should be using (is it genetic sex or one of the others?)
  • Age is set here https://github.com/Nealelab/UK_Biobank_GWAS/blob/master/0.1/00.load_sample_qc_kt.py#L61 as field 21022 (age at recruitment).
  • It is not entirely clear where this comes from in https://github.com/Nealelab/UK_Biobank_GWAS/blob/67289386a851a213f7bb470a3f0f6af95933b041/0.1/00.load_sample_qc_kt.py#L14, but presumably this means the inferred genetic sex field 22001
• Change input to intermediate results following Store post-QC bgen zarr archives for analysis #9
• With https://github.com/pystatgen/sgkit/pull/391 in, add beta values to results and make sure that full computation does not run twice for both that and p values
• Decide whether or not to attempt sample QC filter for "Use 7 standard deviations away from the 1st 6 PCs"
• Add separate MAF threshold for coding variants
• Add tolerance to argmax used in hard call logic (make ties NA)
  • Waiting on https://github.com/pystatgen/sgkit/pull/419
• Group phenotypes to regress together (see https://github.com/Nealelab/UK_Biobank_GWAS/blob/master/imputed-v2-gwas/4_build_pipelines.py#L27)
  • PHESANT output columns might look like 6153_1, 6153_2, 6153_3, ..., 6153_100, 4620 so it is possible to regress some of them together (these subgroupings correspond to one-hot encodings of the data coding for that UKB field)

Note: for details on the Neale Lab repo organizations see Nealelab/UK_Biobank_GWAS#36.