Thread by @cristenw

If you see saddlepoint approximation used in genomics, @weizhouw introduced the concept to fellow @umich grad student @ellenmschmidt ~2012/2013 when we developed GREGOR - a method for assessing significance of enrichment of overlap between a genomic feature and GWAS hits

This allowed us to avoid deep permutation to estimate really small (significant) p-values

SPA was introduced by Daniels 1954, building on the work of steepest descent by Debye, Watson, Darwin (the grandson) & Fowler
projecteuclid.org/journals/annal…

GREGOR ref: csg.sph.umich.edu/GREGOR/

original SPA ref: onlinelibrary.wiley.com/doi/10.1111/st…

SPA was more famously implemented in SAIGE: pubmed.ncbi.nlm.nih.gov/30104761/

Go Wei!! Applying SPA to genomics was an amazing contribution as a graduate student!

Now it’s entirely possible that all the statistical geneticists know about SPA, but I certainly didn’t, so thank you Wei for introducing the concept to me.

My interpretation of the concept is that if you’re trying to see how far away something is (like a small Pvalue), you can climb up to the top of the mountain and estimate if it’s really far, medium far, not so far etc by knowing the height and base of the mountain you’re on.