The Genetic Specificity of Cognitive Tests After Controlling for General Cognitive Ability

[DylanY684]

Behavioral geneticists have identified hundreds of genetic variants that are associated with general intelligence. But what about other cognitive abilities?

A new article by Robert Plomin and his coauthors examined the genetics of scores on cognitive tests, independent of the influence of g. What they found as fascinating.

Combining data from existing datasets, the researchers found that “genomic g” looks a lot like the g observed in test scores. Genomic g accounts for 46.8% of shared genetic variance across 12 tests. This means that genomic g is the major driving force of genetic similarity across test scores–just as regular g is for test score phenotypes.

Where the study gets really interesting is what happens after the authors control for genomic g. In the image below, the correlation matrix on the left shows the raw genetic correlation, and the matrix on the right shows the genetic correlations after controlling for the shared genetic influence of genomic g. After controlling for g, all of the correlations decrease, and some of them even switch from positive to negative!

This means that some genetic variations impact performance across the board (through genomic g). But other variants have more local impacts–and some variants may be associated with higher performance on one test and lower performance on another!

Moreover, some tests are more impacted by genomic g than others–but this relationship is not associated with their factor loading on the genomic g. In other words, this finding is not just an artifact of which tests contribute the most to genomic g.

Another interesting finding was that controlling for genomic g also impacted the genetic correlations with other traits. Generally, these correlations weakened–sometimes to the point of being non-significant. This means that genomic g has an influence on these correlations, but that there is often room for other genetic influences (see example below).

This is a great study which tells us that the genetics of non-g abilities matters. At both the behavioral and genetic level, a full understanding of human cognition requires studying g and narrower mental abilities.

Reposted from X: https://x.com/RiotIQ/status/1916871779680751698

Link to full article: https://link.springer.com/article/10.1007/s10519-025-10213-5

[VeronicaTale]

The finding that SNP heritabilities remained nearly identical before and after g-correction (0.16 vs 0.13) is crucial. This means the total genetic influence on each test is preserved - it’s just the pattern of genetic correlations across tests that changes. The average genetic correlation dropped from 0.45 to near-zero after correction, which quantifies how much of the “positive manifold” in cognitive genetics is due to genomic g versus ability-specific genetic factors. This 46.8% shared genetic variance figure for genomic g is lower than phenotypic g typically accounts for (~50-60%), suggesting genetic influences are somewhat more specific than phenotypic influences.

[Marcelo]

@VeronicaTale What causes the difference between genetic and phenotypic g? Why would they be different?

[VeronicaTale]

@Marcelo Environmental factors likely contribute to the positive manifold in addition to genetics. Things like educational quality, socioeconomic status, nutrition, and general life experiences affect performance across cognitive domains. These environmental factors create additional covariance across tests beyond what genetics alone produces. So phenotypic g reflects both genomic g (46.8%) and shared environmental influences.

[mitchell_odonovan399]

@VeronicaTale

The finding that SNP heritabilities remained nearly identical before and after g-correction (0.16 vs 0.13) is crucial. This means the total genetic influence on each test is preserved - it’s just the pattern of genetic correlations across tests that changes. The average genetic correlation dropped from 0.45 to near-zero after correction, which quantifies how much of the “positive manifold” in cognitive genetics is due to genomic g versus ability-specific genetic factors. This 46.8% shared genetic variance figure for genomic g is lower than phenotypic g typically accounts for (~50-60%), suggesting genetic influences are somewhat more specific than phenotypic influences.

@VeronicaTale While you are correct on the preservation of total SNP heritability, the most radical finding is the nature of the remaining genetic variance. The shift from an average genetic correlation of 0.45 to near-zero, and the emergence of strong negative correlations (e.g., -0.79 for Tower and Spelling), means the total genetic influence is now opposing on certain abilities. This dramatically changes how we conceptualize the genetic architecture of the mind, showing that some variants are trade-offs, promoting one ability while inhibiting another.

[houston.parkkonen139]

@VeronicaTale

The finding that SNP heritabilities remained nearly identical before and after g-correction (0.16 vs 0.13) is crucial. This means the total genetic influence on each test is preserved - it’s just the pattern of genetic correlations across tests that changes. The average genetic correlation dropped from 0.45 to near-zero after correction, which quantifies how much of the “positive manifold” in cognitive genetics is due to genomic g versus ability-specific genetic factors. This 46.8% shared genetic variance figure for genomic g is lower than phenotypic g typically accounts for (~50-60%), suggesting genetic influences are somewhat more specific than phenotypic influences.

The biggest win here is the data. Researchers now have “clean” genetic stats for specific skill, like memory-only or fluid-reasoning-only. That’s a huge step forward. The next round of studies can stop finding genes for “intelligence in general” and start finding genes for “this one specific way I’m smart.”

[jeramy.craven037]

@houston.parkkonen139

The biggest win here is the data. Researchers now have “clean” genetic stats for specific skill, like memory-only or fluid-reasoning-only. That’s a huge step forward. The next round of studies can stop finding genes for “intelligence in general” and start finding genes for “this one specific way I’m smart.”

We shouldn’t jump to conclusions about “genetic trade-offs.” Those strong negative correlations like -0.79 are between abilities that were already weakly correlated to begin with. When you strip out the main genetic booster, the negative residual correlations are just the mathematical consequence of the tests being different, not necessarily proof of truly opposite biological pathways.

[philthegreat]

The post doesn’t clarify that these findings describe broad patterns across thousands of people, not what happens with any individual person. Even though genetic variants are associated with test scores in large groups, they explain less than 10% of why one person scores differently than another. These correlations are interesting patterns worth studying, but they don’t tell us how genes actually work, which specific genes do what, or why your friend might be better at math while you’re better at vocabulary. The pattern is real, but it doesn’t explain the biology very accurately.

[call.me.meg]

Isn’t the 10% figure just the variance explained by current polygenic scores with the limited sample sizes? Twin studies show cognitive abilities are 50-80% heritable, so the genetic influence is much larger. For sure if the methods improve and sample sizes grow, polygenic scores will capture more of that heritability.