Skip to navigation Skip to main content Skip to footer

Machine Learning Analysis of Brain Images, Behaviors and Genotypes to Understand Aging and Mental Disorders

Machine Learning Analysis of Brain Images, Behaviors and Genotypes to Understand Aging and Mental Disorders

Principal Investigator: Dr Jinbo Bi
Approved Research ID: 51296
Approval date: March 3rd 2020

Lay summary

The ultimate goal of this project is to differentiate valid subtypes for two disorders (addiction and depression) that can be homogeneous in clinical manifestation and etiology. As a secondary aim, the project will disseminate validated statistical machine learning models. In the project period of 6 years anticipated, the team will use new scalable methods to integrate genetic markers, neurobiological features, with diagnostic survey variables to derive subtypes. Before the integration, efforts will be made to identify neural biomarkers from imaging derived phenotypes, and derive behavioural features from biosensor data such as from accelerometer. Using the identified features, the team will construct predictive models to predict a specific subtype so hopefully to identify predictors for addiction (or depression) and its subtypes. The team will assess if any genetic variants are associated with a disease subtype. Differentiation of homogeneous subtypes of the disorder, related to short-term or long-term behaviours, co-occurring phenotypes, specific brain structural and functional patterns, will help us identify genetic variation underlying risk for the disorder, improve diagnostic classification, and facilitate clinical decision-making, including matching patients to treatments and targeting prevention strategies. The subtypes, if validated, will help researchers to further personalized treatments of these disorders. The validated machine learning methods and algorithms will pioneer new models for empirical research of disease subtyping. Furthermore, any features that are identified in this project to be indicative of a subtype may help predict differential treatment responses. The features that are predictive of a subtype can help researchers design prevention strategies.