Statisticians crack how to untangle thousands of genetic variants affecting brain anatomy affecting cognition

A new statistical method lets researchers track causal chains through massive datasets where every step is itself high-dimensional. Instead of studying one gene → one brain region → one outcome, researchers can now study hundreds of genes → hundreds of brain regions → multiple cognitive outcomes simultaneously, and measure which genetic-to-brain-to-cognition pathways actually matter. The method works on real data—the Alzheimer's Disease Neuroimaging Initiative dataset with 688 genetic variants, 202 brain regions, and 11 cognitive measures—making it possible to find biological mechanisms in systems too complex to study before.

For decades, neurogeneticists have known that complex traits like memory or cognition involve thousands of genetic variants working through brain anatomy to produce behavior, but they could only study pieces of this chain at a time—one gene at a time, one brain region at a time. The mathematical machinery didn't exist to handle the combinatorial explosion of high-dimensional data at every step. This paper provides that machinery. The practical effect: researchers can now identify which genetic-to-brain-to-cognition pathways are real and reproducible, rather than guessing or studying artificially simplified versions. This matters for Alzheimer's disease and cognitive aging specifically—the Alzheimer's Neuroimaging Initiative application found interpretable disease pathways that older statistical methods would have missed or produced as noise.