Mark ở vũ trụ khác · Another Universe Genes and Rare Diseases #2 · Polygenic inheritance · GWAS

There is no "intelligence gene"

Visscher et al. · American Journal of Human Genetics · 2017

The storythe hunt for an "intelligence gene" and why it failed
Findingcommon traits come from thousands of small genetic effects; rare diseases often from one gene
Evidencegenome-wide association studies (GWAS) in hundreds of thousands to millions of people
LabelSolid; gene scores cannot yet predict an individual

Source: Visscher et al. (2017), American Journal of Human Genetics 101(1): 5–22; Savage et al. (2018), Nature Genetics 50: 912–919.

For reading and understanding, not a substitute for medical advice.

One-minute recap

Essence

12 "intelligence genes" did not hold upRechecked in 2012 in about 9,800 people: 32 tests, only 1 reached the threshold.
Instead: hundreds of regionsA 2018 study of 269,867 people found 205 genomic loci and 1,016 genes, each adding a tiny amount.
Many small genes make a curveHeight: 12,111 variants in 5.4 million people. Added up, they give a bell-shaped distribution.
Rare diseases differMany rare diseases come from one broken gene, like a switch. So they are easier to find: 35% versus 11%.
Genes are oddsA gene score explains 12–16% of differences in education across a population. Not enough to predict one child.

Source: Chabris et al. (2012), Psychological Science 23(11): 1314–1323; Savage et al. (2018), Nature Genetics 50: 912–919; Yengo et al. (2022), Nature 610: 704–712; Smedley et al. (2021), NEJM 385: 1868–1880; Okbay et al. (2022), Nature Genetics 54: 437–449.

Part 1

The question

If cystic fibrosis or PKU is caused by one broken gene, is there a gene that makes us smart, tall or good at maths? Twenty years of reading genomes answer: almost never. Knowing why helps you read "gene test for talent" ads with a clear head.

Question and wording are Mark's, based on Visscher et al. (2017).

2000sHunt one gene"Candidate genes" for intelligence get published.
→
2012RecheckAlmost none hold up.
→
2017–2022Thousands of small genesWhole-genome scans in millions of people.
Part 2

The story

Who, where, when, and in what setting.

2004 · The human genome has "only" 20,000–25,000 genesThe Human Genome Project published a near-finished sequence. Far fewer protein-coding genes than many expected. Few genes, countless traits: they must work together.
2000s · Candidate-gene seasonGroups picked a few plausible genes (dopamine, memory: DRD2, COMT, BDNF, APOE), tested them in hundreds to a few thousand people, and reported genes "linked to intelligence".
2012 · Christopher Chabris checksChabris's team (USA) took 12 published variants and tested them in three samples, about 9,800 people in total. Power calculations predicted 10–15 significant results. Actual: 1 of 32 tests. The title says it plainly: most are probably "false positives".
2017 · Ten years of genome-wide scansPeter Visscher and colleagues (University of Queensland, Australia) reviewed 10 years of GWAS: for almost every complex trait studied, many genomic loci contribute.
2018 · Intelligence: 205 lociA meta-analysis of 269,867 people found 205 loci and 1,016 genes. Only 24 loci were known before. None is an "intelligence gene".
2022 · Height: 12,111 variantsA study of 5.4 million people found 12,111 independent variants spread over about 21% of the genome. The same year, the first truly complete human genome filled the last 8%.

Source: International Human Genome Sequencing Consortium (2004), Nature 431: 931–945; Chabris et al. (2012), Psychological Science 23(11): 1314–1323 (12 variants, 5,571 + 1,759 + 2,441 people, 1/32 tests, 10–15 expected); Visscher et al. (2017), AJHG 101: 5–22; Savage et al. (2018), Nature Genetics 50: 912–919 (269,867 people, 205 loci, 1,016 genes, 24 loci before); Yengo et al. (2022), Nature 610: 704–712 (5.4 million people, 12,111 variants, 21%); Nurk et al. (2022), Science 376 (the missing 8%).

Context: why did single-gene hunts so often go wrong?

With small samples and genes that each add only a little, "significant" results are often luck. Studies with pretty results got published; empty results stayed in the drawer.

Chabris himself said the 2012 results do not mean genes play no role, but that intelligence probably involves thousands of variants, each very small.

Source: Chabris et al. (2012); Chabris quoted in ScienceDaily, Oct 2012 (secondary). The explanation about luck and the "drawer" is Mark's.

Part 3

How genes work, in 3 sentences

1. Many rare diseases are like a switch: one broken gene is enough to cause disease, so reading genes can find the cause.

2. Height, intelligence and education are like thousands of small dials: each variant nudges a little up or down, and only the sum makes a difference.

3. Environment (nutrition, school, family) turns the same dials, so "heritable" describes odds across a group of people, not one person's destiny.

Source: Visscher et al. (2017); Visscher, Hill and Wray (2008), Nature Reviews Genetics 9: 255–266 (heritability is a population parameter); Smedley et al. (2021). The switch and dial images are Mark's.

Two kinds of inheritance. Left: single-gene disease, like a switch: a broken gene means disease, a normal one means none. Right: polygenic trait, thousands of small dials, each nudging a little, adding up to a bell curve. one gene, like a switch common in rare diseases broken genedisease normal geneno disease two separate groups thousands of small genes, like dials height, intelligence, education each dot = one variant, nudging +/− a little low ← add them all up → high
Two kinds of inheritance, two ways to search. Reading genes diagnosed 35% of patients whose disease was likely caused by one gene, but only 11% of those with complex disease. Illustration, no numbers on the axis.Source: Smedley et al. (2021), NEJM 385(20): 1868–1880 (35% and 11%); Visscher et al. (2017) (complex traits involve many loci).
Genetic positions found in large studies: intelligence 205 loci (269,867 people, 2018); education 3,952 variants (about 3 million people, 2022); height 12,111 variants (5.4 million people, 2022). more people, more small genes found intelligence (2018)269,867 people205 loci education (2022)about 3 million people3,952 variants height (2022)5.4 million people12,111 variants
No "height gene" or "intelligence gene", just very many small ones. Bar length is proportional to the count. The counts measure slightly different things (loci, independent variants), so compare only the scale.Source: Savage et al. (2018), Nature Genetics 50: 912–919; Okbay et al. (2022), Nature Genetics 54: 437–449; Yengo et al. (2022), Nature 610: 704–712.
Part 4

Fact check: facts and myths

13 things often said about genes and intelligence, checked against the original research. Tap a row for the reason.

Share of differences explained by gene scores. Height in European-ancestry people: 40%. Height in other ancestries: about 10 to 20%. Years of schooling: 12 to 16%. how much of the differences do gene scores explain? 100% height, European ancestry40% height, other ancestries10–20% years of schooling12–16%
Most of the differences lie outside the gene score. And scores built from European-ancestry data work much worse for people of other ancestry, Vietnamese people included.Source: Yengo et al. (2022), Nature 610: 704–712 (40%; 10–20%); Okbay et al. (2022), Nature Genetics 54: 437–449 (12–16%).
SOLIDIntelligence is partly heritable

Decades of twin studies agree (Plomin and Deary 2015, Molecular Psychiatry 20: 98–108).

SOLIDMost complex traits involve very many genes

Ten-year GWAS review (Visscher et al. 2017).

SOLIDHeight is linked to more than 12,000 genetic variants

12,111 independent variants, 5.4 million people (Yengo et al. 2022).

SOLIDHumans have only about 20,000–25,000 protein-coding genes

International Human Genome Sequencing Consortium (2004), Nature 431: 931–945.

SOLIDSingle-gene diseases are easier to pin down than complex ones

Whole-genome sequencing diagnosed 35% versus 11% (Smedley et al. 2021).

CAUTION"Intelligence is up to 80% heritable"

Plomin and Deary (2015): about 20% in infancy, rising to "perhaps 80%" in later adulthood, mostly from twin studies in rich countries. The figure shifts with age and population.

CAUTION"A gene score tells you how well your child will do at school"

The score explains 12–16% of the variation in years of schooling across a population; comparing siblings, the direct effect is only about half (Okbay et al. 2022).

CAUTION"Gene scores work equally well for every ancestry"

Close to false: the height score explains 40% in European-ancestry people, only about 10–20% in others (Yengo et al. 2022).

FALSE"There is an intelligence gene"

205 loci, 1,016 genes, each contributing very little (Savage et al. 2018).

FALSE"The published 'intelligence genes' were confirmed"

Rechecking 12 variants: 1 of 32 tests reached the threshold (Chabris et al. 2012).

FALSE"50% heritable means half of your intelligence comes from genes"

Heritability describes differences within a group, it does not split one person in two (Visscher, Hill and Wray 2008).

FALSE"Choosing embryos by gene score gives a much smarter child"

Simulation: on average only about 2.5 IQ points, with wide error (Karavani et al. 2019, Cell). The European Society of Human Genetics (2022) called it "unproven, unethical"; the American College of Medical Genetics and Genomics (2024) does not recommend it.

FALSE"Genes are destiny"

Even in a single-gene disease like PKU, the right diet from birth changes the outcome completely (GeneReviews, "Phenylalanine Hydroxylase Deficiency"). For polygenic traits, environment plays an even bigger part.

Source: as noted in each row; full list in Part 9.

Part 5

Where we are today

From gene maps to a market for "choosing children".

2019 · Simulating embryo selectionShai Carmi's group (Israel) calculated that picking the top-scoring embryo over an average one gains only about 2.5 cm of height and about 2.5 IQ points, with very wide error. In large families, the child with the top height score is usually not the tallest.
2019 · In VietnamNewspapers reported services "decoding genes to find prodigies and talents" in children, and experts warning they lack scientific basis.
2022 · Two "saturated" mapsHeight: 12,111 variants, capturing nearly all the heritability from common variants in European-ancestry people. Education: 3 million people, 3,952 variants, explaining 12–16%.
2022 · Europe speaksEuropean Society of Human Genetics: using polygenic scores to select embryos is "unproven, unethical".
2024 · The US speaksAmerican College of Medical Genetics and Genomics: insufficient evidence of clinical utility; not recommended as a clinical service.

Source: Karavani et al. (2019), Cell 179; Tuổi Trẻ 28 Dec 2019, Báo Pháp Luật (secondary); Yengo et al. (2022), Nature 610: 704–712; Okbay et al. (2022), Nature Genetics 54: 437–449; Forzano et al., ESHG (2022), European Journal of Human Genetics 30(5): 493–495; Grebe et al., ACMG (2024), Genetics in Medicine 26(4).

Part 6

Verdict

Scores are Mark's assessment based on the sources in Part 9, not figures from a study.

Soundness of "polygenic"
5/5
Helps understand rare disease
4/5
Predicting one person
1/5
Fair across ancestries
2/5
Worth remembering
  • Science corrected itself: single-gene hunts failed, whole-genome scans took over
  • Samples of millions, results that replicate
  • Explains why rare diseases are easier to trace
Keep in mind
  • Gene scores are weak at predicting a person
  • Data mostly from European-ancestry people
  • Used to sell "talent selection" services

Label: SOLID. There is no "intelligence gene"; there are thousands of small ones plus environment, and they speak of odds across a group rather than one person's fate.

More: why is this piece in a "rare diseases" series?

Because it is the other side of the coin. Rare diseases often come from one gene with a big effect, so reading genes can point straight at the cause. Common traits come from thousands of genes with small effects, so reading genes only gives odds. Understand both and you can read any genetic test result correctly.

The "More" box is Mark's comment, based on sourced facts in Parts 2 to 5.

Part 7

Lesson: genes are odds, not destiny

Four questions before trusting a "gene test for talent" ad.

For reading and understanding, not a substitute for medical advice.

Four questions: how many genes; what percentage explained; tested on Asians yet; about a group or one person. 1 · how many genes?a few = suspiciousintelligence: hundreds of loci 2 · explains how much?ask for the %education: 12–16% 3 · tested on whom?on Asians yet?weak outside Europe 4 · group or person?odds ≠ destinycannot predict one child A vague answer to any of these is a signal to stop.
Each question rests on a sourced number.Source: Savage et al. (2018) (205 loci); Okbay et al. (2022) (12–16%); Yengo et al. (2022) (much weaker outside European ancestry); Visscher, Hill and Wray (2008). The four questions are Mark's suggestions.
Count the genesA service "reads 10 talent genes"? Intelligence involves hundreds of loci, each adding very little.
Ask for the percentage"How much of the variation does it explain?" If nobody can answer, it is advertising.
Ask "tested on whom?"Scores built from European-ancestry data work much worse for other ancestries.
Think in oddsRare single-gene disease: ask a clinical geneticist. Polygenic traits: upbringing and learning are still most of the story.
Self-check: read a gene-test ad in 5 minutes

1. Find an ad for a "talent gene test" or "sports gene test".

2. Underline how many genes they read. Fewer than a few hundred for a trait like intelligence: suspicious.

3. Look for "how much variation it explains". None: write "no answer".

4. Check which country's people the original data came from.

5. Count the promises about "one child" (your child will...). Each one overstates.

The self-check is Mark's suggestion, not a study protocol.

Part 8

Read next

Visscher et al. (2017), "10 Years of GWAS Discovery"American Journal of Human Genetics 101. Free on PubMed Central. The most readable overview of genome-wide studies.
Chabris et al. (2012), "Most reported genetic associations with general intelligence are probably false positives"Psychological Science 23. A model of how to recheck old results.
Genes and Rare Diseases #01: Photo 51DNA as a string of letters, and the story of who found its shape. Read
Genes and Rare Diseases #03: Huntington and Lake MaracaiboA disease caused by exactly one gene, and the woman who drew a family tree of thousands to find it.
Part 9

Sources

Main sources are original peer-reviewed papers and statements from genetics societies. The working environment blocks PubMed and journal sites, so figures were checked against those pages' own excerpts in search results (looked up 16 Oct 2026), without opening full texts. Secondary sources are marked.

  1. Visscher PM, Wray NR, Zhang Q, Sklar P, McCarthy MI, Brown MA, Yang J (2017), "10 Years of GWAS Discovery: Biology, Function, and Translation", American Journal of Human Genetics 101(1): 5–22. PubMed Central
  2. Chabris CF et al. (2012), "Most reported genetic associations with general intelligence are probably false positives", Psychological Science 23(11): 1314–1323.
  3. Savage JE et al. (2018), "Genome-wide association meta-analysis in 269,867 individuals identifies new genetic and functional links to intelligence", Nature Genetics 50(7): 912–919. PubMed
  4. Yengo L et al. (2022), "A saturated map of common genetic variants associated with human height", Nature 610: 704–712.
  5. Okbay A et al. (2022), "Polygenic prediction of educational attainment within and between families from genome-wide association analyses in 3 million individuals", Nature Genetics 54(4): 437–449.
  6. Plomin R, Deary IJ (2015), "Genetics and intelligence differences: five special findings", Molecular Psychiatry 20(1): 98–108. PubMed
  7. Visscher PM, Hill WG, Wray NR (2008), "Heritability in the genomics era: concepts and misconceptions", Nature Reviews Genetics 9: 255–266.
  8. International Human Genome Sequencing Consortium (2004), "Finishing the euchromatic sequence of the human genome", Nature 431: 931–945. PubMed
  9. Nurk S et al. (2022), "The complete sequence of a human genome", Science 376.
  10. Smedley D et al. (2021), "100,000 Genomes Pilot on Rare-Disease Diagnosis in Health Care – Preliminary Report", NEJM 385(20): 1868–1880.
  11. Karavani E et al. (2019), "Screening Human Embryos for Polygenic Traits Has Limited Utility", Cell 179.
  12. Forzano F et al., on behalf of ESHG (2022), "The use of polygenic risk scores in pre-implantation genetic testing: an unproven, unethical practice", European Journal of Human Genetics 30(5): 493–495. PubMed Central
  13. Grebe TA et al. (2024), "Clinical utility of polygenic risk scores for embryo selection: A points to consider statement of the ACMG", Genetics in Medicine 26(4).
  14. GeneReviews, "Phenylalanine Hydroxylase Deficiency" (NCBI Bookshelf).
  15. Secondary sources: ScienceDaily, Oct 2012 (Chabris quote); Tuổi Trẻ, 28 Dec 2019, "Giải mã gen xác định 'thần đồng': Tin khoa học hay tin con mình?"; Báo Pháp Luật, "Thực hư xét nghiệm gen tìm tài năng".