environMENTAL Methodological approaches for MRI data integration biomarker discovery and validation within the environmental project

In this talk, I will give an overview of the analytical strategy we will employ for biomarker discovery in the environMENTAL project. To find generalisable biomarkers to predict and stratify mental disorders, we will need to solve many methodological challenges including how to meaningfully aligning data from heterogeneous cohorts spanning the whole lifespan, fusing data from modalities with very different characteristics, accounting for complex patterns of missing data and extracting generalisable and interpretable low dimensional representations from complex datasets.
Moreover, many of these analyses need to be performed in a decentralised and distributed manner. I will give an overview of some of the analytical techniques that we will employ to solve these challenges, including federated machine learning techniques, normative modelling, deep learning, transfer learning and classical penalised multivariate regression techniques. I will illustrate by discussing in detail how such techniques can be applied to neuroimaging data but it should be remembered that they are all more widely applicable.