To what extent does brain structure correlate with different psychological traits? An interesting new paper from Massachusetts General Hospital researchers Mert R. Sabuncu and colleagues uses a new method to examine what the authors call the ‘morphometricity’ of various behaviours and mental disorders.
Sabuncu et al. define morphometricity as “the proportion of phenotypic variation that can be explained by macroscopic brain morphology”–in other words, the degree to which people with similar brains tend to be similar in a particular behaviour. Morphometricity is somewhat analagous to the concept of heritability from genetics.
The results showed that Alzheimer’s disease is almost perfectly morphometric, with an estimated value of 0.94–1.00 (where possible values range from 0 to 1). Schizophrenia was moderately morphometric (estimate 0.55), with autism coming in slightly lower at 0.38. Perhaps surprisingly, Parkinson’s disease had a much lower morphometric value of just 0.20.
Other, non-disease-related traits, such as IQ and level of education, were highly morphometric too, with values above 0.8. In fact, IQ was slightly more morphometric than sex (IQ 0.95 vs. sex 0.93), while age was perfectly morphometric (1.00).
The authors conclude that
In the dawning era of large-scale datasets comprising traits across a broad phenotypic spectrum, morphometricity will be critical in prioritizing and characterizing behavioral, cognitive, and clinical phenotypes based on their neuroanatomical signatures. Furthermore, the proposed framework will be significant in dissecting the functional, morphological, and molecular underpinnings of different traits.
This is an important paper, but we shouldn’t rush to over-interpret the results. For instance, whereas Sabuncu et al. say that the high morphometricity estimates for disorders such as autism and schizophrenia“unequivocally point to a neuroanatomical substrate for these clinical conditions”, this really doesn’t follow.
We should not conclude that high morphometricity means that brain structure causes a particular behaviour.
In terms of the morphometricity measure itself, I note that the authors say that it “does not require cross-validation, which is often the technique used in machine learning to gauge prediction accuracy”. That’s because their method “exploits the entire dataset to fit the model and estimate the unknown variance component parameters, and in turn morphometricity, in an unbiased fashion.” I would prefer if some validation had been performed, e.g. by calculating the morphometricity of a randomly generated ‘trait’ or by permuting the trait data.