iiinova — where your digital self keeps evolving

BTW, I'm always looking for students, postdocs, collaborators and minions. DM me if you're interested in working together.

011

(and apologies that the peer review process took so bloody long...)

120

This project could not have been done without the mentorship of @gs2747.bsky.social & @jkpritch.bsky.social and the hard work of @hakha.bsky.social , Julie Zhu and @courtsmithrun.bsky.social.

130

Side note: As part of the prolonged review process, we showed (in our supplement) using extensive data analysis and simulations that while COJO hits are not necessarily causal, they do a phenomenally good job at tagging the number, frequency and effect sizes of the true underlying causal variants.

131

Our conclusion is that the genetic architecture is well-described by a model of pleiotropic stabilizing selection, and well-approximated by a single distribution of selection coefficients for all traits. Differences between traits are driven by scaling with target size and heritability per site.

However, after we scale effect sizes by the heritability per site and account for differences in GWAS power, the genetic architectures of height and FEV1 look identical. The same is true for all other traits as well.

The same isn’t true of traits that differ in their heritability per site, like height and FEV1.

100

Therefore, two traits that differ in their target size but not in their heritability per site will differ only in the number of variants affecting them, but not in the variants’ joint distribution of frequencies and effect sizes. Just what we see for height and platelet crit.

121

The number of variants affecting a trait is proportional to the target size. The squared effect size of these variants (in units of the phenotypic variance) is proportional to the heritability per site, the heritability over the target size.

So why do traits differ in their genetic architecture? While the distribution of selection coefficients is similar between traits, traits vastly differ in their target size and heritability. The genetic architecture scales with these two parameters:

111

As validation of our inferred distribution of selection coefficients we looked at allele ages: RELATE infers the GWAS hits for our 95 traits to be younger than matched controls, indicating they are under selection. Our model predicts very well the distribution of allele ages.

The single shared distribution (or SSD) model fits the data very well and much better than simple heuristic models with a Normal distribution of effect sizes.

120

We therefore, suggest a useful approximation where we assume that there is a single shared distribution of selection coefficients among traits.

100

We infer these 3 components for 95 continuous traits in the UK biobank. While there are differences in the distribution of selection coefficients between traits, their confidence intervals overlap.

100

Our model has three components: (1) The target size for a trait - the number of sites where a mutation would affect a given trait. (2) The distribution of selection coefficients at those sites. (3) The mean heritability per site – the heritability divided by the target size.

100

For example, in the UK biobank, there are approximately 1500 independent GWAS hits for height which explain about 40% of height’s heritability. For FEV1, there are only 350 hits that explain roughly 10% of the heritability. How can we explain such differences?

100

Even using the same dataset, GWAS for different traits identify different number of significantly-associated genetic variants (“GWAS hits”) for different traits and these variants explain different proportions of the traits’ heritabilities.

100

Simple scaling laws control the genetic architectures of human complex traits

Why do complex traits differ in their genetic architecture? In our new PLOS Biology paper, we will try to convince you that two simple scaling laws drive differences in the number, effect sizes and frequencies of causal variants affecting complex traits. Thread: journals.plos.org/plosbiology/...

journals.plos.org

Genome-wide association studies have revealed that the genetic architectures of complex traits vary widely. This study shows that differences in architectures of highly polygenic traits arise mainly f...

18738

This project could not have been done without the mentorship of @gs2747.bsky.social & @jkpritch.bsky.social and the hard work of @hakha.bsky.social, Julie Zhu and @courtsmithrun.bsky.social.

100

100

100

100

The same isn’t true of traits that differ in their heritability per site, like height and FEV1.

100

Stanford University Press @stanfordpress.bsky.social · 30/09/2025

100

Reposted by Yuval Simons

Population Biology Modeling & Theory (PBMT) is a peer-reviewed journal reporting advances in modeling and theory within population biology. Its scope spans demography, ecology, epidemiology, evolutionary biology, population genetics, and phylogenetics. PBMT will be online soon.

14514

Reposted by Yuval Simons

Laura K Hayward @lkhayward.bsky.social · 24/08/2025

Why do males and females often differ in traits? The expected answer: selection. But our new paper in GENETICS shows that genetic drift alone can generate sexual dimorphism — even when male & female optima are the same

411049

Reposted by Yuval Simons

Yoav Gilad @ygilad.bsky.social · 20/06/2025

I heard in the elevator a graduate student joking about destroying their lab and starting over, and I was just thinking that the lab I did my PhD. in was actually completely destroyed this week.

051

Reposted by Yuval Simons

Genetics Society of America @genetics-gsa.bsky.social · 03/06/2025

Jennifer Blanc and @jeremyjberg.bsky.social use theory and simulations to study the process of testing for an association between polygenic scores and axes of ancestry variation when confounding factors are present. Learn more about their findings in #GENETICS: buff.ly/EmKpXyP

01710

Yuval Simons @yuvalsim.bsky.social · 25/04/2025

One year on the admissions committee and I've become vehemently pro GRE. Without it we end up prioritizing professional resume padders and missing students of non-standard backgrounds.

0101

Our work suggests a novel approach for modeling epistasis using rank statistics and provides a new resampling approach to ascertain statistical significance. We hope it finds use across multiple disciplines.

010

However, in a third system (GB1-IgG binding, Olson, Wu and Sun 2014) our method struggles because of the interplay of mutation effects on both protein binding and folding, suggesting the need to expand our work to consider two (or more) underlying scales of effect.

We see similar results for the PDZ-CRIPT system.

100

For FOS-JUN, we see that position that are in close physical distance are highly enriched for specific (non-global) epistatic interactions and, vice versa, that the interactions we identify correspond to positions in close physical contact.

100