Thank you for coming to the meeting and for your great talk!

We’re excited to share this work and hope it proves useful across GWAS, PRS, heritability, and environmental confounding studies.

Massive congratulations to @RoohyShemirani and our whole team for seeing this through.

Feedback and collaborations welcome! 🧵👇

9/9

Method is open-source & scalable 🚀
➡️ Python package: github.com/roohy/spc
➡️ Simulation framework: github.com/roohy/msprim...

8/9

Why does this matter?
As we move into diverse biobanks and rare variant discovery, we will need more methods. SPCs offer a modern toolkit for accurate, scalable correction of recent demographic structure.

7/9

Environmental confounding?
SPCs dramatically improved adjustment:
📉 Reduced Moran’s I more than PCs
📊 Avoided false heritability from geography
🧬 Preserved signal from true polygenic traits

6/9

Rare variant GWAS?
In UKBB inflation factor dropped from 7.6 (PCs) to 1.2 (SPCs) — helpful for uncovering real associations instead of noise.

5/9

In simulations, SPCs captured >90% of fine-scale population structure with fewer components than PCs, which explained <50%.

In GWAS of environmental phenotypes, SPCs reduced genomic inflation by 12%, while matching PC performance for highly heritable traits like height.

4/9

SPCs offer a novel solution:
✔️ Based on IBD graphs, not allele frequencies
✔️ Transforms local IBD sharing into continuous covariates using graph spectral theory
✔️ Works out-of-the-box with large datasets (tested on N ≈ 420K from UK Biobank)

3/9

Population structure is a major confounder in GWAS and heritability estimation.
Principal Components (PCs) have been the default — but they fall short for recent, non-linear, fine-scale structure, especially in rare variant analyses.

2/9

Oh no, I’m so sorry to hear this. Condolences to his family and friends.

A conclave of scientist ..

A conclave of scientist ..

Thanks Sarah! 💛

Thank Simone!!! 💛

Thanks Joe! 💛

🌍 Hillerød, Denmark from May 19-23, 2025,

🚨 Abstract submission closes March 15, 2025—still a few spots left! Submit yours here: sciencecluster.dk/event/genomic-mechanisms-of-disease-2025