Project
Linking brain function and structure to phenotypes: does more data in fixed sample size lead to higher replicability?

Brain-wide association studies are defined as ‘studies of the associations between common inter-individual variability in human brain structure/function and cognition or psychiatric symptomatology’ (Marek et al., 2021). To find these associations one generally requires thousands of brain voxels or regions and commonly employes correlation methods. However, an approach known as multivariate BWAS is preferred to integrate all these features across different brain regions into a single predictive model for a specific phenotype to improve power and reliability (Woo, Chang & Lindquist, 2017; Bzdok & Steyerberg, 2021). A problem concerned with BWAS is that finding replicable brain-behaviour associations requires thousands of individuals. Thus, to find the same results with the same methods for new data one would need to increase the sample size (Botvinik-Nezer & Wager, 2023). Moreover, the statistical methods usually employed for multivariate BWAS use noisy data but do not handle the measurement error properly (Elliott et al., 2020; Gell et al., 2023). To accommodate this, new statistical analysis methods should be employed that consider that brain features and phenotypes are not measured without error.

The aim of this PhD is to develop a break-through in these problems by developing and testing new statistical procedures that use more information and accommodate measurement error. The hypothesis is that using more data from each participant leads to improved replicability and secondly that latent variable models for BWAS also improves replicability. Hence, it is not necessarily the sample size but the number of data points that needs to be increased to detect a small, possibly noisy effect. In this study we will develop nuclear norm penalized latent variable models (Principal Covariate Regression and Reduced Rank Regression) with multiple phenotypes for brain-wide association studies and test whether these new approaches lead to increased replicability. We will develop these new statistical methods for BWAS, assess replicability in large-scale simulation studies and further test the methods on existing cohort data.

Supervisors
Prof. Dr. Mark de Rooij
Dr. Wouter Weeda

Financed by
NWO

Period
25 January 2025 – 25 January 2030