Related work from TOPMed

[hammer]

Stephanie Gogarten of UW has a few repos worth perusing, though she uses R and not Python: https://github.com/UW-GAC/analysis_pipeline shares similar goals to this repository but for TOPMed data; https://github.com/UW-GAC/GENESIS and https://github.com/smgogarten/GWASTools have code to support the TOPMed pipeline.

[eric-czech]

A few notes:

• The QC part of the pipeline is similar to the UKBB QC process
• A brief overview of the steps:
  • LD pruning to remove uncorrelated variants (via SNPRelate)
  • Kinship estimation robust to pop. structure (via SNPRelate implementation of KING-robust)
  • Estimates above provided to PC-AiR for PCA robust to relatedness, but not pop. structure (via GENESIS)
  • It is interesting that PLINK offers "Relatedness Pruning" as an explicit step that PC-AiR embeds within the preprocessing steps (i.e. it runs PCA on a subset of unrelated samples)
  • PCA robust to population structure using PC-Relate (via GENESIS)
  • This requires the PCA vectors from the PC-AiR step
  • The scaling strategy for this involves operating first on groups of samples (+ all variants) and then operating on pairs of results from sample blocks. That's a strategy I had thought about before but haven't seen implemented anywhere until this
  • Association testing using kinship estimates as random effects and PCs as fixed effects (via GENESIS)
  • Visualization beyond usual QQ/manhattan plots (via LocusZoom)
• SNPRelate + GWASTools + GENESIS together all make for a pretty comprehensive toolkit