FEMA-Long framework models unstructured covariances to detect time-dependent genetic effects at scale

Parekh, Parker, Pecheva, Frei, Vaudel, and colleagues across institutions including the University of Oslo, University of California San Diego, and the University of Oxford have published a methods paper in PLOS Genetics introducing FEMA-Long — an extension of existing mixed-effects modelling frameworks designed to handle unstructured covariance matrices in large-scale longitudinal genomic datasets.

Standard linear mixed-effects (LME) models used in genome-wide association studies (GWAS) typically assume random intercepts or simple stationary covariance structures over time. FEMA-Long relaxes these assumptions, allowing the model to capture time-varying genetic effects and more complex within-individual correlation patterns. The authors demonstrate the method using data from large neuroimaging and developmental cohorts, identifying time-dependent associations that simpler models would not have detected.

The work addresses a practical challenge in biobank-scale longitudinal research: as studies such as the UK Biobank and cohorts in the ABCD Study collect repeated measurements over years, the statistical infrastructure for analysing these data needs to keep pace. FEMA-Long is described as computationally tractable at scale, which the authors argue will make it applicable across imaging genetics, paediatric development studies, and other longitudinal domains where repeated measures are now routine.

The paper was published in PLOS Genetics on 11 June 2026 and represents a methods contribution rather than a discovery of specific biological variants.