@jonatanpallesen
If the apparent placebo depends mostly on a tendency to regress to a healthy mean, there should be no placebo effect in RCTs of terminal cancer patients, or Alzheimer's disease, or other diseases that follow a predictable deterioration path. Is that true?
In PNAS today, Greg Clark documents the inheritance of social status in an impressive dataset on 422,374 people born in England between 1600 to 2022.
Clark finds a strong persistence in social status going from close to distant relatives, which fits a model dating back to RA
New educational attainment (EA) GWAS out today: . We expand the sample size from ~1 million to ~3 million, making it one of the largest GWAS to date. We identify more loci affecting EA and increase our ability to predict EA from genetic data, and more...🧵
Noah is someone I know personally from Oxford. I don't agree with his politics or some of the conclusions of his research, but I know he is an intelligent and open minded truth seeker. If there is no place for heterodox scholars in the academy, that's very worrying.
I have just learned that the mob has been successful in getting social scientist Noah Carl fired from his position at Cambridge. The statement from St. Edmund's apologises for the "hurt" caused by his appointment. Read the back-story here
Out today in
@NatureGenet
, our paper showing how to increase power for family-based analyses by imputing missing parental genotypes: . We apply our software () to 9 phenotypes, demonstrating confounding in standard GWAS and more..🧵
Impressive work finds 8 genes with a large effect on educational attainment in UKB whole exome data. Mouse knockouts of one of the genes, Kdm5b, show impaired memory. The people who said genomic studies of education would never reveal any meaningful biology were wrong.
New preprint! We studied how rare protein-coding variants impact cognitive function & linked diseases in
@uk_biobank
. We identified 8 large-effect cog fx genes, dosage-sensitivity for KDM5B, additivity of rare variant and polygenic risk & more (1/12)
One can add the throttling of academic freedom as another factor contributing to the exodus to industry. Why accept a lower salary and worse conditions if you can't research what you want or speak freely?
The discussion around Greg Clark's paper () on the inheritance of social status has centered on whether models other than a genetic model can explain observed relative correlations.
It has been known since the work of Cloninger, Rice, and Reich in the
Out today in
@ScienceMagazine
, our paper on cross-trait assortative mating and why estimates of genetic correlations (i.e. shared genetic effects on phenotypes) are likely inflated. See this thread from the preprint for an overview of the idea and results
New from Richard Border
@andywdahl
@flint
Noah Zaitlen myself & others. Estimated genetic correlations between traits may be inflated by cross-trait assortative mating (xAM): . So are apparent genetic relationships between traits a statistical artefact? 🧵
I wrote an article for Nature Genetics () giving my view on the newly published work measuring heritability using whole-genome data. WGS data will give us a more complete picture of the sources of heritable variation than ever before. 🧵
This study quantifies the overall contributions of both common and rare variants to the phenotypic variation in a complex trait (height or BMI in this case),
Comprehensive analysis of dominance (non-additive) effects in UK Biobank () finds negligible non-additive variance from common variants. However, it also finds that variants with large effects or phenotypes whose genetic architecture is dominated by a small
My paper is out! . Assortative mating (AM) occurs when parents' phenotypes are correlated. My paper aims to solve problems AM creates for estimation of indirect genetic effects (IGEs) and heritability by developing theory and an estimation method...🧵
Out today in
@NatureGenet
: our paper on RDR, a novel method for estimating heritability without environment bias (). I wrote an accessible blog post explaining the paper and its implications ()
#genetics
New from Richard Border
@andywdahl
@flint
Noah Zaitlen myself & others. Estimated genetic correlations between traits may be inflated by cross-trait assortative mating (xAM): . So are apparent genetic relationships between traits a statistical artefact? 🧵
I'm a reviewing editor at eLife, and I'm opposed to removing
@mbeisen
for his social media posts. It is bad for scientific culture when the expression of unpopular opinions is policed with extreme strictness. This is true whether those opinions are thought of as 'right-wing' or
If you are a tenured professor in the life/neuro sciences who values open exchange of ideas in academia, please sign our new petition telling HHMI and eLife not to censure Michael Eisen for tweeting about the Middle East.
@eLife
and
@HHMINews
Further thoughts on Greg Clark's recent work in PNAS.
The paper has two major contributions:
1) Construction of a large genealogical database on 422,374 people born in England between 1600 and 2022 along with measures of social status — occupational status, higher education
In PNAS today, Greg Clark documents the inheritance of social status in an impressive dataset on 422,374 people born in England between 1600 to 2022.
Clark finds a strong persistence in social status going from close to distant relatives, which fits a model dating back to RA
@mattyglesias
This is known as the problem of missing heritability. What we can account for by GWAS is not the same as the total heritability, but some argue that twin estimates are still too high. I wrote an accessible article discussing these issues:
In
@TrendsGenetics
today,
@hilsomartin
and I deliver a spotlight () on the recent exome-wide association study of education and cognitive ability. Our concluding paragraph states:
The study by Chen et al. provides a compelling rebuke to those who have
New blog on 'missing heritability' and the preprint on estimating heritability from whole genome sequence (). Complements my appearance on the
@insight_podcast
with
@razibkhan
(). Missing heritability not yet solved.
#genetics
John Carlisle, an editor of Anaesthesia, found that 25% of randomized controlled trials (RCTs) had data that was impossible to trust when examined at the individual participant level, but only 1% had such obvious flaws in the summary data reported in the journal article.
People
Some researchers say that at least a quarter of randomized clinical trials in some fields are untrustworthy (perhaps fake?!), but still cited in reviews that guide medical practice.
How big is this problem & what's the evidence? Here's my look in
@Nature
My manuscript on estimating heritability without environmental bias ( ) and Augie Kong's manuscript on the nature of nurture ( ) out now on
@biorxivpreprint
. Heritability and genetic association may not be as they seem.
#Genetics
Our review article on 'Deconstructing the sources of genotype-phenotype association in humans' out today in
@sciencemagazine
(). We lay out the contributions to genotype-phenotype associations, and the value of family data for separating them.
@molly_przew
To accompany
@carl_veller
and
@Graham_Coop
's recent article in
@PLOSBiology
, I wrote an accessible primer on family-based genome-wide association studies:
Genome-wide association studies (GWASs) can be affected by confounding. Family-based GWAS uses random, within-family genetic
3.4 million person GWAS of smoking and alcohol use: . Polygenic predictors can now explain 10% of the variation in smoking initiation. An example of the utility of behavioural GWAS for biomedicine: genes influence smoking behaviour and therefore health
My opinion piece on the missing heritability problem, what we know about it, and the challenges any solution to it would have to overcome.
#genetics
#Genomics
Large-scale genomic analysis of metabolic syndrome highlights the brain as the most relevant tissue. BMI and metabolic traits should be considered as partly behavioural traits. The wall between 'behaviour genetics' and 'medical genetics' is much thinner than most suppose.
Very cool ancient DNA study on bronze age kefir from Xinjiang.
I suspect that fermentation techniques - which reduce lactose - explain why lactase persistence didn't evolve rapidly across populations consuming dairy.
Now we know more about the cultures and techniques of
Mendelian randomisation can be led astray by population stratification and assortative mating. This paper shows how genetic data on close relatives can be used to remedy this:
Explainer on new work (with Augustine Kong, the SSGAC, and others) detailing our method for imputing parental genotypes from sibling and parent-offspring data () to increase power for family based analyses. Software here () (1/n).
Out today in Behavioral and Brain Sciences,
@DamienMorris
,
@StuartJRitchie
, and my response to
@callie_h_burt
's critique of the use of genetic predictors in social science research: .
We conclude: Burt is correct that social scientists should include
@razibkhan
I wonder how many of the academics speaking out against the removal of Eisen would do so for a right-wing academic who said something offensive to left wing orthodoxy. My guess: almost none of them.
My paper with
@genemodeller
on finding genetic effects on phenotypic variability with application to BMI in UK Biobank out today in
@NatureGenet
. Short accessible blog here: . Software freely available here:
A lot of people made (spurious) criticisms of plots showing years of education (EA) against our predictor, the EA4 PGI, due to the lumpy distribution of EA. So here's a bonus unpublished plot with a less lumpy outcome variable, Average English & Maths GCSE grades from MCS. 🧵
Genetic effects that influence where you live can induce gene-environment correlations that confound GWASs. We show that controlling for geography reduces heritability for SES and reduces genetic correlations with SES for many traits.
Our new preprint:
If you flip a coin twice, the chance you get heads at least once is 50%+50%=100%. If you flip a coin three times, it would be 150%, but that is a whole other thing…
This is greatly overstated. While BMI is an imperfect measure of adiposity, it is often the only measure available in many datasets. This is because height and weight are nearly always measured and things like waist circumference are not. Should we throw all of that data out? No
My request for 2023: stop using BMI.
If you're a member of the public, using it to suggest an 'ideal' weight: STOP.
If you're a clinician, using it to judge someone's weight: STOP.
If you're a scientist, using it as a proxy for adiposity: STOP.
#StopUsingBMI
#EpiTwitter
New work from Richard Border and others (including myself and
@andywdahl
) that develops a computationally efficient software package, xftsim, for forward simulation of complex intergenerational dynamics including both genetic and cultural transmission from parents to offspring.
As descendants of early Neolithic farmers who spread into Europe from the near east, the Sardinians' ancestors have been in Europe for thousands of years longer than the Sami. But only the Sami are indigenous? This idea is an American import that doesn't work in Europe.
Today is
#SamiNationalDay
!
The Sami are the EU’s only indigenous people. Since time immemorial, they have lived in an area stretching across parts of four countries: Finland, Norway, Sweden and Russia. This area is called Sápmi.
New work led by
@OmanGuan
, and presented this morning at
#ASHG22
, on novel estimators for family-based GWAS that increase power and robustness: . We introduce two novel estimators that maximize power for estimating direct effects in different situations. 🧵
Our recent work on imputing missing parental genotypes for family-based genetic association analyses has been featured as a research highlight in Nature Methods, featuring an interview with myself and Augie Kong.
Research Highlight by Lin Tang: researchers describe a statistical method that imputes parental genotypes for accurate estimation of direct genetic effects driving genotype-phenotype associations.
@NatureGenet
paper:
Highlight:
Thanks for a nice writeup
@SashaGusevPosts
!
A few comments:
1. "Though, it is worth noting, the family GWAS can also introduce some biases by effectively only testing the children of heterozygous parents at each variant (see: [Veller, Przeworski, Coop, (2024)] for more), or
I wrote a bit about the two very interesting studies of siblings/families from last week. Tan et al. family GWAS () and Sidorenko et al. sibling heritability estimates (). A few surprising findings summarized here: 🧵
Feldman states "there’s no reason to think these statistical correlations have any causal meaning", referring to correlations between education and education PGS. However, Feldman wrote a paper showing within-family analyses can recover causality () (1/4)
Further thoughts:
The paper has two major contributions:
1) Construction of a large genealogical database on 422,374 people born in England between 1600 and 2022 along with measures of social status — occupational status, higher education status, literacy, dwelling value,
Critics of education GWAS say common variants have tiny effects and no obvious biological function. Point them to genes that have large effects, and they'll say they're just mutation intolerant genes that affect everything. Is there any type of genetic variant they like?
Happy to announce that I have started a new position at UCLA with
@Dan_J_Benjamin
at the Anderson School of Management and the Human Genetics department.
@ent3c
@Quillette
So you're saying that several articles that have found a detectable decline in the average polygenic score, including one I was involved in published in PNAS and covered in the Guardian (), are fringe science?
Final version of my opinion piece on 'Solving the missing heritability problem' is in
@PLOSGenetics
. Here's a related podcast I did hosted by
@razibkhan
:
A great example of how a socially mediated behavior (smoking) is affected by genetic variation. The variants that disrupt CHRNB2 protect against smoking and therefore against emphysema, COPD, and lung cancer. The boundary between behavioral and biomedical genetics is a construct
Thrilled to share our human genetics discovery of the protective association between rare variants in CHRNB2 (encoding a subunit of nicotinic acetylcholine receptors) and smoking addiction, now out in
@NatureGenet
. See the reply for a detailed thread. Below is a short summary.
I read this paper by
@_twolfram
and
@DamienMorris
in more detail recently. It's a worthwhile read for anyone interesting in intergenerational educational inequalities, (missing) heritability, and assortative mating.
The paper applies an extended twin and family design (ETFD)
My paper with
@_twolfram
is out in NPJ Science of Learning! We find that estimates of shared environmental influence for educational attainment in conventional twin studies may exaggerate the differences in educational opportunity between families: 🧵1/21
This is a bad decision by
@eLife
. To police the political opinions of editors to such an extreme degree will have a further chilling effect on a scientific culture that is already broken when it comes to free and open debate.
I have been informed that I am being replaced as the Editor in Chief of
@eLife
for retweeting a
@TheOnion
piece that calls out indifference to the lives of Palestinian civilians.
I was looking at my Google Scholar (sad, I know) and I noticed that there is a rough inverse relationship between the effort I put into a paper and the citation rate. My most highly cited paper was a blog post that got turned into a paper. Has anyone else noticed this?
This editorial () proposes to "refuse publication of (or retract post-publication)" on grounds that are so vague and subjective they could be argued to apply to almost anything, including...(1/2)
This is a nice paper (that I reviewed) that shows how an econometrics method, ORIV (obviously related instrumental variables), can be used to adjust for measurement error in analyses of genetic predictors (PGIs).
PGIs are computed from GWAS summary statistics, which estimate
An important analysis that shows the association between (non-transmitted) parental genotype and offspring education is perhaps not primarily due to 'genetic nurture', and is at least in part due to population structure/assortative mating.
Unless you stop people from accessing their genomic data, regulating pgs will be impossible. There will be web tools hosted outside the US where one can upload one's data to compute pgs. All regulation will achieve is increased costs to providers based in the US.
A proposal that the FDA should regulate direct-to-consumer delivery of polygenic scores.
In my opinion this is not convincing. What is the evidence that consumers are harmed? Why should this be a priority for the FDA and worthy of their efforts?
With
@razibkhan
, I discuss heritability, the genetics of education, and some of the criticisms made of behavior genetics in the wake of
@kph3k
book, The Genetic Lottery.
From January, I will be working in LA with the SSGAC (). We're hiring a statistical genetics postdoc (). Come work with me,
@Dan_J_Benjamin
,
@patrickaturley
, and others at the frontier of social science genetics. Please retweet!
Exciting new work from
@qinqin_huang
@EmilieWigdor
@hilsomartin
and others (myself included): 'Dissecting the contribution of common variants to risk of rare neurodevelopmental conditions'. 🧵
While it is understood that both rare mutations (often de-novo) and common variants
1/ We’re excited to share our new preprint “Dissecting the contribution of common variants to risk of rare neurodevelopmental conditions” available on medRxiv!
@jew_AndAhalfMen
That's certainly part of it but I think attitudes to alcohol have changed too. Perhaps part of it is not wanting photos of you as a drunken mess appearing on social media.
@SashaGusevPosts
@charlesmurray
If you're using the pgs as a measure of total genetic effect, that's not valid reasoning. I'm not sure where else you get the less than 20% number from.
Thoughtful review of
@kph3k
’s The Genetic Lottery. There’s much I agree with, but I would give more credit to genomic methods in strengthening the argument that genetic factors play a causal role in social inequalities and that it is not all ‘red hair gene’ type effects (1/2).
Heritability of longevity could be lower than believed on the basis of twin studies and analysis of pedigrees (). Great to see models including both assortative mating and cultural transmission applied to such a huge pedigree dataset.
Thanks to
@SashaGusevPosts
, we noticed an error in the publicly available summary statistics for age at first birth, where only the first four chromosomes were available. This has now been fixed and the full summary statistics for age at first birth are available to download.
I wrote a blog post that is relevant to the discussion on non-additive effects: . Basically, most variance will be additive unless you have a non-monotonic genotype phenotype map (e.g. overdominance) that leads to zero or very small additive effects. This
Comprehensive analysis of dominance (non-additive) effects in UK Biobank () finds negligible non-additive variance from common variants. However, it also finds that variants with large effects or phenotypes whose genetic architecture is dominated by a small
Would be curious to see a within-family analysis using siblings done on the results of this paper. That would test the contribution of population stratification, indirect genetic effects, and assortative mating -- all likely to be substantial. Plenty of siblings in UKB.
Nice post explaining bias in GWAS effect estimates. I would add that an important source of 'genetic stratification' is assortative mating, which can lead to potentially very large bias in marginal effect estimates. (1/2)
@cremieuxrecueil
@NatureGenet
I think it's maybe harsh to blame the reviewers since this is a tricky technical issue that no one realized was potentially a problem until me, as far as I'm aware.
The issue is that people got used to doing standard gwas and using low quality imputed genotypes is fine for
Interesting work furthering the argument that stratification needs to be taken more seriously in human genetics than simply using PCs or mixed models. Aligns with my recent anlyses of family data, which are consistent with a substantial contribution from stratification to GWAS.
We (with
@mathiesoniain
) have been thinking a lot about the effects of population stratification, especially on polygenic scores over the last year. Thrilled to finally share some findings (TL; DR):
I agree that EA displays the most confounding issues out of any well-studied phenotype. However, the difference with other traits is one of degree, not kind. Social science genetics has taken the issue of confounding seriously and studied it extensively. Other subfields often
A 🧵 on educational attainment (EA) GWAS and evidence of extensive confounding from multiple recent studies (much more than for disease traits). FWIW: I’m not a behavioral geneticist and I welcome disagreement on this.
A great thread on the FTO locus, which harbors a common variant with a strong effect on BMI and obesity. My take on this: GWAS reveals meaningful human biology, but that biology can be complex and take a lot of effort to work out, and it may be quite different in humans than in
Buckle up! We're in for a wild ride today. A new
@NatMetabolism
paper by scientists from China adds a surprising twist to the long-known FTO GWAS story.
The FTO locus (16q12.2) is the first ever GWAS locus to be associated with obesity and even after 16 yrs now, scientists
The exome sequencing plus imputation pipeline developed by Regeneron is impressive in its efficiency for discovering genetic associations that are potential drug targets.
But there is something to be said for having the complete genome sequences of all participants in a biobank
Whole genome sequencing (WGS) and whole exome sequencing (WES) have championed genomic discovery, but how do they compare? Published in
@Nature
, our team analyzed ~150K
@uk_biobank
participants to study WGS vs. WES, and their impacts on discovery in genetic association studies.
Interesting results on 'genetic nurture' and how to interpret the effects of polygenic scores: . Estimates of direct effects of educational attainment, cognitive ability, and depression polygenic scores less than half the size of the standard estimates.
If you're at all interested in leaving academia post-PhD, every paper you write should have a corresponding github repo. Super simple way of demonstrating to future employers you actually know how to do things
A polygenic index is a weighted sum of many SNP genotypes, so will be approximately normally distributed in general (provided the weights are somewhat evenly spread). This doesn't automatically imply anything about the phenotype distribution in the population.
This is not merely an attack on social science genetics - it’s a dismissal of genetics in general as a hopelessly subjective endeavor. And when traits are normally distributed - well that’s what those bad eugenicists thought!
As detailed in our review paper (), standard GWAS estimates of SNP effects include contributions four distinct sources: the direct effect of inheriting an allele, indirect effects from relatives, population stratification, and assortative mating (2/n).
Important results on stratification and its effects on both measures of selection and LD-score regression. LD-scores and population differentiation found to be correlated (likely due to background selection), casting doubt on it as a clean method for assessing stratification.
It has always been my intuition that individuals or small groups are more likely to produce truly novel science than large teams. This study confirms this. I can't help but think of these huge GWAS consortia that do useful but rarely truly novel work
Excited to be moderating/facilitating a session at ASHG 2019 in Houston on 'Potential policy implications of genetic research on education' with four brilliant speakers:
@kph3k
,
@daltonconley
,
@melindacmills
, and
@AysuOkbay
. Also excited about the tacos and BBQ.
A subtle point that this paper makes concrete is that confounding in the GWAS used to make a PGS complicates the interpretation of within-family estimates of PGS associations across traits: i.e., just because PGS of trait 1 predicts trait 2 within-family, that doesn't imply
New preprint by
@carl_veller
and myself on the interpretation of population and family-based genome-wide association studies in the presence of confounding
@SpencrGreenberg
My article on the missing heritability problem casts some light on twin estimates of heritability - as discussed here - in relation to more recent genomic data based estimates:
@SashaGusevPosts
@charlesmurray
You are ignoring decades of twin and behaviour genetics research, though, and the substantial uncertainty in the RDR estimate as well as the fact SNP heritability isn't total heritability.
I am inclined to think that twin studies have overestimated heritability, but to