Using Connectomics and Machine Learning to Examine Brain Development

Session Type: Paper Symposium

Abstract

The field of connectomics, or the study of the brain’s structural and functional connections, is gaining popularity as a framework to study brain development. Along with connectomics, the use of machine learning and predictive modeling to ‘predict’ developmental outcomes from brain connectivity data is growing. Predictive models, rather than explanatory models, provide the opportunity to identify novel treatment targets and identify individualized interventions for neurodevelopmental and psychiatric disorders. In this symposium, four studies of brain development at the intersection of connectomics and machine learning across infancy, childhood, and adolescence. In the first study, functional connectomes from infants and toddlers age 8-24 months (Baby Connectome Project) are used to reliably predict infant chronological age. The second study, also focusing on infants and toddlers, presents several studies at the forefront of using connectomics and machine learning to aid in early identification and intervention for neurodevelopmental disorders. In school-aged children, the third study uses data from the ABCD study with functional connectomes and predictive modeling to examine functional connections that are predictive of G (g-factor, general intelligence). The fourth study uses connectome-based predictive modeling in a sample of adolescents to predict irritability from functional connectomes during “frustration” task. Overall, this symposium will highlight the utility of combining connectomics and machine learning (predictive modeling) for the study of brain development across infancy, childhood, and adolescence. Discussions after each study is presented will include best practices for predictive modeling of developmental outcomes for brain-based models and future directions in this area of research.

Sub Unit

• Biological Processes: Neuroscience and Genetics

Chair

• Alexander Dufford, Northwestern University

Individual Presentations

• Building connectome-based predictive models in infancy and toddlerhood - Presenting Author: Monica Rosenberg, University of Chicago

• Infant neuroimaging and prediction: towards early identification and individualized intervention - Presenting Author: Jessica Girault, University of North Carolina at Chapel Hill; Non-Presenting Author: Mark D. Shen, University of North Carolina at Chapel Hill; Non-Presenting Author: Martin Styner, University of North Carolina at Chapel Hill; Non-Presenting Author: Joseph Piven, University of North Carolina at Chapel Hill; Non-Presenting Author: John R. Pruett, Washington University in St. Louis

• The Architecture of Neurocognitive Abilities in the Developing Functional Connectome - Presenting Author: Chandra Sripada

• Functional Connectivity during Frustration: Predictive Modeling of Irritability in Children and Adolescents - Presenting Author: Wan-Ling Tseng, Yale School of Medicine; Non-Presenting Author: Eva McAdam Freud, Yale School of Medicine; Non-Presenting Author: Dustin Scheinost, Yale School of Medicine