Intro to Precision Medicine

Some of these ideas come from a grant that Satra shared called Risk Predictors of Mental Health through Multimodal Biotyping Intelligent Assistive Framework for Mental Health Management

What is precision medicine?

• Precision medicine is the approach of predicting and diagnosing mental health disorders using:
  • microbiome, inflammation, genomics, neural markers (fMRI, MRI, EEG, pupillometry), behavior, clinical symptoms

What is the overarching goal of pHealth?

Improved risk assessment and early detection of mental illness.

How will this goal be achieved?

Our aim is to develop biologically driven risk predictors and monitors of mental health. Using these datasets, we will relate genetic information to clinical evaluations through endophenotypes that span brain imaging, cognitive and behavioral testing, and information from wearable devices and speech recordings.

Developing these kinds of models are not without significant challenges but the potential outcome is enormous: providing useful brain measures, and new categories of disorders.

What are the short term goals?

• To find biotypes based on combining genetic and brain imaging data
• To develop risk predictors that integrate biological and behavioral information.
• How will these goals be achieved?
• To test these models, we would deploy them in research clinics.
• The outcome of this work will feed into a more longitudinal goal of capturing real-world behavior and interventions in order to build models to monitor and deliver interventions that are tuned to individual trajectories.

What is the long term goal?

The long term goal is to create an intelligent assistive framework for mental health management that supports integration of multiple sources of live and longitudinal data, learns phenotypic associations and builds interpretable, predictive models, and suggests individualized actions towards targeted goals.

What is the problem?

• Clinical assessment is resource intensive and inaccessible to those who need it most. When accessible, initial assessments are ineffective: do not consider societal and community options to boost wellbeing.
• Why are ML approaches often ineffective?
  • ML diagnostic approaches often focus on specific disorders rather than symptoms.
  • ML approaches are often devoid of connections to underlying biology

What is the opportunity?

• The advent and growth of consortial data collections that include genomic assays, clinical and behavioral data together with brain imaging offer a unique opportunity to understand genotypic and phenotypic variation in a large population.
• Datasets:
  • UK Biobank, adolescent brain cognitive development study, genus dataset, healthy brain networks

Considerations:

• Holistic approaches:
  • societal, cultural, community-based interventions in conjunction with (instead of) ML-guided healthcare
• Ethics:
  • data anonymity? Who do the tools serve? Are they biased against certain groups?
  • how is model uncertainty communicated to scientists and non-scientists alike?