This is a new conceptual figure which explains how to make our proofs. I hope this idea is helpful to other population geneticists.

You can also run a fast scan with randomized phenotypes to double check if selection is confounding.

Sadly, the test is prone to confounding due to very strong positive and recent selection (see LCT in European ancestry cohorts). You can automatically check for this as a part of the automated workflow.

Two of the six genome-wide significant signals are about current therapeutic targets, e.g., at the NBAS gene.

I'm excited to present my work on rigorous and scalable multiple-testing corrections for complex haplotype scans in WGS at ASA STATGEN 2025!
t.co/rv205OBCs1

Beyond the Temple and Thompson asymptotics, empirically the IBD rates look ~= normally distributed, outside of many the selected loci

The assumption of IBD rates having exponentially decaying correlations has good empirical evidence across ancestry groups and simulations. We're thinking about if an IBD segment covers both positions, and the segment lengths are often modeled as exponential/Gamma.

The main reason we wrote this is to have a GW significance level for an African ancestry analysis, whereas everyone already knows about LCT selection in Europeans. We replicated a few signals across TOPMed and UKBB samples. I find the chr16 hit compelling.

You can run my software github.com/sdtemple/isw... with a few command lines and a cluster. We did this for many ancestry groups. The LCT, MHC, and OCA2 signals are shared across European and Indian cohorts.

In hard sweeps, the test has good power if the selection coefficient exceeds 0.01. Some consider this very strong selection, but w/o multiple testing. We propose sweep detection in a hypothesis testing framework, as opposed to as a prediction task.

Family-wise error rate is reasonably well controlled. And the GW significance level is much larger than the Bonferroni method, so you can have some power gains. When FWER is inflated, this can be understood by the Temple and Thompson conditions.

Think of the paper as a bunch of correlated t tests. The Y notation below is the IBD rate, and mu is the genome average IBD rate.