Thread by @antonioregalado

Our story today on DNA tests that predict intelligence is generating lots of feedback. Here's a THREAD to go with it.

2/ First of all here's the story in @techreview . As scientists zero in on hundreds of genetic links to high or low IQ one result will be PERSONAL, DIRECT TO CONSUMER, DNA IQ TESTS technologyreview.com/s/610339/dna-t…

3/ This image is an actual report on predicting IQ (versus everyone else) you can get online. I generated it by uploading a 23andMe "variant" file to @Gn_Plz . They have a $4 "Intelligence App" for sale that does the calculation.

4/ Worth saying right now this is not based on my DNA. It belongs to a guy in Florida who has his 23andMe file published at @OpenHumansOrg . So, before you put your DNA online, keep in mind people could profile you in strange and new ways, getting in your head.

5/ Intelligence isn't a property of your body, like, say, the length of your big toe. But it is something that can be measured. And if it can be measured then geneticists can search for genetic differences between people that predict it.

6/such predictions can operate in complete ignorance of the biological basis of intelligence, or whatever trait you might be looking at. they're called "polygenic risk scores".

See our 2018 Breakthrough list for a description of the tech technologyreview.com/s/610251/forec…

7/ Naturally, there is no single "gene for intelligence." It's a trait that's the result of the entire organism, the fact that we have a brain, etc. So really every gene is involved.

8/The DNA sequence of human genes varies between people. If you measure a lot of people's DNA and then give them an IQ test, you can look for variations linked to the test result. Each variant has a teeny tiny effect. But you can add them up into a "score". A polygenic score.

9/ Genetic studies are getting larger and larger. They now sometimes involve >1 million people. That's the key to finding the predictive genetic variants.

Record size study was 1.3 million looking for genetic clues to insomnia. technologyreview.com/s/610151/a-sea…

10/ Powering these studies are resources like 23andMe and especially the UK Biobank, which last summer open sourced the DNA data of 500,000 people plus ALL KINDS of stuff measured about them. 2,000 traits, i think.

Here's a story about that. technologyreview.com/s/609184/uk-bi…

11/ What it means is anyone (anyone who is a geneticist with a computer) can start modelling the genetics of ANY trait.

It's all just going directly online. Here's just one "heredity browser"
nealelab.github.io/UKBB_ldsc/h2_b…

12/ Twin studies put the inherited part of intelligence at ~ 50%. Funny is that genetics predicts intelligence better the older you get. its like 80% late in life. Maybe because you of u shape your environment, pick your friends, etc, reinforcing the genes.

13/ I am being corrected by @razibkhan to say there is a heritable component to intelligence, not that it's inherited. This is a nuance that's hard to grok but important. Maybe razib will blog about it.

14/ In January, Robert Plomin and Sophie von Stumm published a review in Nature called "The New Genetics of Intelligence". Huge eye opener vis a vis how how growing size of studies is leading to prediction of intelligence from DNA. nature.com/articles/nrg.2…

Currently, measured DNA differences can explain about 5% of the differences in IQ and 10% of educational attainment, among large groups of people. So it doesnt explain all of the heritable part. And none of the environmental part, i guess.

16/ Here's a box that explains how the predictions are limited by (1) what is genetic (2) the SNP genotyping data used (3) and by the fact that the studies, though big, are not yet big enough.

17/ By the way Robert Plomin, who is 70, and runs a big twin study at King's College, was ranked among the 100 top psychologists of the entire 20th Century, including dead ones.

18/ Someone compared Plomin's interest in polygenic scores to "like when Apple incorporates a news technology in the iphone." The big boys are playing with it.

Worth noting when Plomin's career began many denied any role for heritability in intelligence.

19/ Others didn't want anyone to find the genes. In 1996, activists calling themselves the “Campaign for Real Intelligence” called on Britain’s Medical Research Council to revoke Plomin’s funding, saying genetic discoveries “can only lead to elitism” newscientist.com/article/mg1502…

20/ At first no one could find any genes linked to IQ. Studies were too small. That changed in 2017 with a Dutch report that found 22 genes. @carlzimmer wrote about it here calling it an "enormous success" nytimes.com/2017/05/22/sci…

21/ Now, a year later, rumor is there is a paper pending in Nature Genetics that looks at the genes of more than 269,000 test takers and finds variants implication 1,000 genes!! Another study of education involves 1 million people

22/ Here you can see a plot of the predictive ability of DNA improving, with bigger study sizes. So, marching into the future, we'll predict better.

23/ Here are my own plots showing the data for height, also strongly genetic. Geneticists are basically trying to use DNA to explain where people land on that "bell curve." The more people you study, the more of the trait you can explain.

24/ These sort of polygenic predictions are ALREADY in consumer DNA tests, like the diet ones which predict body mass index. 23andMe told me that 16 out of their 21 trait reports are based on polygenic scores. Here's the list and an example (photonic sneeze)

25/ So last year I began wondering, how come there aren't any genetic reports on IQ or educational attainment (a proxy for IQ)? I mean, the predictions for intelligence are BETTER than some of the predictions 23andMe gives people.

26/ At first, 23andMe said, oh we never really thought of doing an intelligence report . Yeah, righhhhht. Because 23andMe is heavily involved in intelligence gene hunts. They contribute a ton of data.

27/ Sadly, 23andMe never opened up to tell me the story of their internal deliberations over this sensitive question. They provided the following statement though. It's very interesting.

28/ But it turns out a couple of other genetic app stores DO calculate I.Q. scores for anyone, including GenePlaza.com . Also DNA.Land but founder @erlichya says he doesn't think the scores predict well at all. medium.com/@dl1dl1/can-we…

29/ Telling people their predicted I.Q. is a crappy business proposition because half of people will be below average. Not the kind of news people pay to hear.

30/ DNA.Land thinks its own I.Q. scores are basically worthless. But look at the REAL reactions of people who score low. It's fatalism or denial. I tweeted them earlier. See:

External Tweet loading...
If nothing shows, it may have been deleted
by @antonioregalado view original on Twitter

31/ So I did get in touch with MENSA the "high IQ society." It's a club you have to ace an I.Q test to join. If you score > 132 you can join MENSA and get discounts at Avis etc.

32/ So, Mensa had never heard of genetic I.Q. scores. But a spokesman told me they would be "very interested" in them. Here's is the statement they have me.

33/ FREE BUSINESS IDEA: I really think some consumer gene test entrepreneur needs to get with MENSA right now and develop a branded DNA IQ app.

34/ What is serious is what if DNA scores are used to identify predicted smart kids or dumb ones, and then make educational choices based on that. Plomin calls it "precision education". He's real fuzzy on the details of what that means tho. Sounds pretty GATTACA

35/ Here's a bunch people with their DNA prediction versus their actual score on the GCSE, a UK test people take at 16. Plomin looks at this and says wow, these predictions really work. Others would say the opposite!

36/ In general the DNA score correlates with the test score. But check out the outliers. Some people with smart genes did bad on the test, and vice versa.

37 There can and will be a big debate over whether, when, and if people should generate this DNA information about kids. Here's Plomin saying it's OK "so long at the probabilistic nature of the prediction is kept in mind." Because we are all so good at understanding probability.

38/ the counterpoints to Plomin's advocacy of testing kids's DNA can be found at Gloomy Prospect, the blog of @ent3c Eric Turkheimer. See "The Ethics of GPS" and several others. geneticshumanagency.org/gha/blog/

Fav quote. as fair as "predicting your IQ from a cousin you haven’t met.”

39/ Now we get to the eugenics part. A potential killer use case for polygenic scoring, as inaccurate as it is, is to decide WHO GETS BORN. That is, you could pick the IVF embryo with the highest predicted IQ, or perhaps the lowest chance of mental illness, if that's you concern

40/ Last November, we wrote about a company Genomic Prediction, that is trying to implement polygenic scoring systems in IVF (not easy). One of the founders, Steve Hsu, is on the record wanting to select smarter people. technologyreview.com/s/609204/eugen…

41/ END. Look for another thread soon discussing David Reich's race editorial in the New York Times in the context of polygenic DNA scores for intelligence.

Correction to this thread: Carl Zimmer didn’t call the 2017 IQ study an ‘enormous success’ in the NYTIMES. That phrase in the headline was a quote from a researcher. Carl, whose careful moderation is renown, called it a “significant advance” .