Predicting individual brain structural health from the individual expotype: the role of the internal expotype

Promoting brain health is key to alleviate burden on healthcare systems and economies in the years ahead. Grey matter health as measured with structural MRI in an aging population has been shown to relate to a myriad of different factors spanning biomedical, environmental, life style, socio-affective and early life factors. However, an holistic view on the whole exposome, that is considering a wide range of factors together, is missing in that context. To address this gap, we trained machine learning algorithms to predict individual grey matter health in the UK BioBank population based on more than 200 variables spanning different exposome domains. To generate a whole-brain normative value of grey matter health for each participant, we first computed the Brain Age Gap (BAG) which reflects the gap between the expected age based on grey matter pattern and the chronological (true) age of the participant. We then trained a random forest model to predict this computed gap for any individual participant based on the individual expotype (261 variables).
When tested in new participants (our-of-sample predictions), the model could predict grey matter health (BAG) to a reasonable extend (r = 0.23). Using a SHAP explainer to gain more insight into the model then revealed that the large spectrum of considered factors contributed to the predictions. However, factors pertaining to cardiovascular health and related life style factors (including diet and smoking), diabetes and bone health consistently show greater contribution to the prediction across different subsets and algorithms. These findings highlight the role of individual expotype diversity, mainly related to the cardiovascular, metabolic and musculoskeletal systems, in accelerated grey matter aging. This may enable early identification of individuals at higher risk of unhealthy brain aging and inform public health strategies.