Brain age gap (BAG) is a well-established scientific concept using MRI imaging data and prediction of BAG using machine learning methods have been reported by many groups. In our earlier work, we used transfer learning and vision models to predict BAG and associate that with genetics in UK Biobank using ~ 30K volunteers’ MRI images.
In this scientific project, as the first goal, we are proposing to utilize our previous experience and refine our machine learning models for predicting brain age gap for a larger cohort (~90K) volunteers’ MRI images and associate that with common and rare variants for exploring the genetic architecture of brain aging and finding genetic signals of resilience and susceptibility to neuropsychiatric and neurodegeneration conditions. If the data permits, we also want to explore potential differences in BAG prediction based on genetic ancestry.
As a second goal, we wish to utilize deep learning to refine machine learning models for predicting brain aging using the plasma proteomics data generated by the plasma proteomics project we participated and compare that to MRI based brain age prediction results.
As a third goal, we wish to explore how metabolome-based biological age gap prediction method correlates with proteome-based age prediction as well as investigate if the metabolome-based biological age gap prediction results are better correlated with our refined brain age gap prediction methodology based on MRI images and proteome.
Finally, genetics associations with predicted brain age will be explored for all three modalities separately and combinatorially for common and rare variants using genetics data (WES, WGS). Associations with prescription drug and co-morbidity patterns will be also explored.
This project will run for 3 years and we hypothesize that by using advanced transformer-based AI methods to integrate these three diverse modalities our understanding of incidence and progression of diseases of the CNS can be advanced.
