New preprint!. We unveil the profound connection between the Ancestral Recombination Graph and Linear Mixed Models. Guess what? Random effects are merely a consequence of mutations running on a fixed genealogy.

RT @bookeditor_: 허준이 직접 쓴 동의보감 "초고본"이 발견되었다는 놀라운 소식. 이번 "초고본"은 1613년 찍힌 "초간본"을 출간하기 위해 허준이 직접 집필한 최초의 원고로 보인다고. 와우. 🔍기사 읽기. https://t.c….

RT @AJHGNews: 📣New from Tang & @CharlestonCWKC!.📄A genealogy-based approach for revealing ancestry-specific structures in admixed populatio….

RT @FabrizioRomano: 🚨❤️🤍 BREAKING: Viktor Gyökeres to Arsenal, here we go! Verbal agreement in place between all parties involved. Sportin….

RT @ErnestRyu: Two cents on AI getting International Math Olympiad (IMO) Gold, from a mathematician. Background:.Last year, Google DeepMin….

RT @MarcusxRashford: Hello culers.👋🏾💙❤️

RT @NatureRevGenet: New online! An evolutionary continuum between non-coding and coding DNA

RT @SxrgioSZN: 😂😂😂

RT @sokrypton:

RT @umichkim: Predicting microbial community dynamics is hard: mechanistic models lack flexibility, ML models overfit & ignore physics. Thi….

RT @zeyun_lu: Happy to share that our work from the @nmancuso_ lab is out in @NatureGenet! We developed SuShiE, a multiancestry fine-mappin….

RT @nocontextfooty:

The remaining Q. is that to what extent the new goalpost, the mutational variance, is relevant to your empirical question. I do think that this mutational quantity is somewhat irrelevant to the nature vs nurture debate that's popular in this website. 5/5.

Hence, the criticism against GCTA, LDSC etc. for not accounting for LD due to assuming independent random effects goes away if you move the goalpost from estimating the usual genetic variance w.r.t individual sampling to mutational variance w.r.t. the ARG. 4/n.

We compute LD covariance via Cov(G_p, G_q) where p and q are two different loci. This is zero once you condition on the ARG. Sounds surprising? it's mathematically obvious that P(X) anf P(X|Y) have different mean, variance, etc. 3/n.

ARG-LMM suggests that at least linkage disequilibrium (LD) due to linkage can be safely removed by conditioning on the ARG. ARG contains all recombination information, so conditioning on it removes all variance due to LD. 2/n.

What I can tell is that standard h2 estimators are mainly capturing genic variance, which is the sum of individual locus variance over all loci. As many papers have suggested, this ignores various forms of disequilibria (e.g. LD) caused by AM/selection etc. 1/n.

RT @ErnestRyu: 7. However, LLMs will become exceedingly powerful for problems that *someone* knows how to solve (in-distribution, in traini….

Long story short, genomic h2 ignores everything else except mutation and short generation times in family designs effectively suppresses mutation's contribution. Hence, two are capturing disjoint genetic signals.

When you say *genetic* variance, you should first stipulate what genetics you're talking about. That could be mutation, drift, recombination, mating or whatever forces that shapes the genetic landscape.

Based on my recent preprint, I think the missing heritability or the gap btw genomic and family studies is a feature and not a bug. The former measures mutational variance over a long period of time while the latter quantifies the mendelian variance over few generations.