| Title: | Replicability of multivariate brain-behaviour associations depends on clinical profile |
| Journal: | Communications Biology |
| Published: | 25 Jun 2026 |
| Pubmed: | https://pubmed.ncbi.nlm.nih.gov/42350741/ |
| DOI: | https://doi.org/10.1038/s42003-026-10364-z |
| URL: | https://www.nature.com/articles/s42003-026-10364-z_reference.pdf |
Publication 20530
WARNING: the interactive features of this website use CSS3, which your browser does not support. To use the full features of this website, please update your browser.
Abstract
Recent work suggests that thousands of individuals are required in multivariate brain-behaviour analyses to obtain consistently replicable results. Some believe, however, that smaller sample sizes may be sufficient if specific subpopulations are targeted. We investigate how sample size and cohort composition influence the replicability of Canonical Correlation Analysis (CCA) results using the UK Biobank (N = 40,514). We apply CCA to diffusion-weighted magnetic resonance imaging (dMRI) phenotypes and cognitive assessment test scores. We define four participant cohorts based on clinical profile and find that, across all cohorts, sample sizes of ≈500 are needed to obtain replicable canonical correlations and variable loadings. The most targeted cohort (comprising individuals with a history of psychoactive substance use) requires much fewer samples to achieve similar or greater correlations than the other cohorts. Our findings support the idea that moderate sample sizes from targeted cohorts can be sufficient for obtaining replicable brain-behaviour associations.</p>
6 Authors
- Michelle Wang
- Brent C. McPherson
- Bratislav Misic
- Franco Pestilli
- Celia M. T. Greenwood
- Jean-Baptiste Poline
1 Application
| Application ID | Title |
|---|---|
| 45551 | Mechanistic modelling of neurodevelopment, neuroplasticity, and neurodegeneration |
Enabling scientific discoveries that improve human health