This project will use the UK Biobank imaging and behavioral data to identify robust mental health brain biotypes, and to assess the extent that these biotypes can be predicted from behavioral and survey data alone.
First, we will derive functional brain signatures from the UK Biobank resting-state and task-evoked fMRI datasets, focusing on large-scale mental health-related networks such as the central executive, default mode, and salience networks. We will characterize individual differences in connectivity patterns and activation contrasts, summarizing them as network-level features.
Second, we will integrate these imaging-derived features with a rich set of behavioral measures, including cognitive test performance, mental health questionnaires (PHQ-9, GAD-7, RDS-4, etc.), lifestyle factors, and actigraphy-derived activity and sleep measures. Using multivariate and clustering methods, we will define latent subgroups (biotypes) and/or continuous scores that jointly capture variation across brain and behavior.
Third, we will evaluate the reproducibility and external validity of these biotypes by testing their associations with clinical outcomes. We will then train predictive models to determine whether these biotypes can be reliably inferred from behavioral features alone, without imaging input. This will test the feasibility of scalable, low-cost prediction of brain-based phenotypes from widely available behavioral and survey data.
The project aims to advance our understanding of how brain network connectivity relates to mental health outcomes, and to establish behavioral proxies for imaging-based biotypes that could inform scalable risk stratification in population and clinical settings.