Pathfinder

One approach, in which the Madrid team under Roberto Colom have been prominent, is to cast the assessment in the form of a computer game. This makes it accessible to a much wider audience, and increases the number of wider-ranging intelligence test which can be used as part of genetic studies of intelligence. In fact, the “gamification” is rather light, a surface gloss only, but it seems to have been enough.

Now Robert Plomin’s team have added further to their own already published game-based intelligence test, and have interesting new results to report. The author’s names are a good roll-call of the new wave of intelligence researchers.

https://www.biorxiv.org/content/10.1101/2021.02.10.430571v2

They point out that genome-wide association studies haven’t yet explained as much variance as can be obtained from twin studies. Critics have called this “missing heritability”, which misses the point. We know from twin studies that intelligence is heritable, and genome-wide association studies are trying to identify the genes responsible for this result. (We know that genetics is powerful in real life, now we need to show it in theory). Part of the problem is that larger studies have put together results from disparate tests, so the team has designed a 40 item intelligence game which produces a reliable (internal consistency = .78, two week retest reliability = .88) measure of g which they have given to 4,751 young adults from their twin study.

This novel g measure, which also yields reliable verbal and nonverbal scores, correlated substantially with standard measures of g collected at previous ages (r ranging from .42 at age 7 to .57 at age 16). Pathfinder showed substantial twin heritability (.57, 95% CIs = .43, .68) and SNP heritability (.37, 95% CIs = .04, .70). A polygenic score computed from GWA studies of five cognitive and educational traits accounted for 12% of the variation in g, the strongest DNA-based prediction of g to date. Widespread use of this engaging new measure will advance research not only in genomics but throughout the biological, medical, and behavioural sciences.

So, they have verbal and nonverbal scores, and can generate a gene-based prediction which accounts for 12% of intelligence variance, a good result by current standards. The test used item response theory to select the most powerful items, and get maximum predictive power from the fewest number of items.

The software is free, and if it gets taken up by further genome wide association studies then the variance accounted for may rise from 12% to 30%, assuming that the heterogeneity of test is a complicating factor which universal use of this test would overcome. The great advantage of a polygenic risk score for intelligence is that you can get a prediction from birth onwards, which overcomes the problem that early tests of intelligence do not get reliable till about age 11, and gain in accuracy thereafter. Early predictions might be a more precise way of evaluating whether early teaching has any effect on later intelligence.

To my eye the missing letter test doesn’t seem worth including. Sure, it is a basic process measure, but a bit out on a limb in factorial terms.

This very big sample of 4,751 25-year-olds shows significant sex differences in favour of men. The authors don’t comment on this, but it fits the emerging pattern of a male intellectual advantage in adulthood.

They say:

Heritability was 57%, shared environmental influence was 8% and multivariate polygenic scores predicted up to 12% of the variance. The latter finding –that 12% of the variance of Pathfinder g can be predicted by DNA –makes this the strongest polygenic score predictor of g reported to date. Although 12% is only one fifth of the twin study estimate of heritability, we hope that adding Pathfinder g in large biobanks will improve the yield of 21meta-analytic GWAS analyses by increasing sample sizes and decreasing heterogeneity of cognitive measures. It should be possible to use the brute force method of increasing sample sizes, especially with less heterogeneity of measures, to close the missing heritability gap from 12% to the SNP heritability of about 30%.

Getting above 30% of the variance will be hard, to reach the 60% heritability revealed by twin studies. That will require whole genome sequencing.

The summary is that the team have created a good new 15-minute IQ test which correlates well with the many longer assessments used over the years on their very large sample of twins. It also has good predictive power. If more widely adopted, and the few bits of explanatory English language translated into other languages, it could be a very useful contribution to large GWAS investigation of the genetic basis of intelligence.