Uncovering functional connectivity patterns predictive of cognition in youth using interpretable predictive modeling.

Area of Science:

• Neuroscience
• Cognitive Science
• Data Science

Background:

• Brain-wide association studies (BWAS) using functional MRI (fMRI) link brain function to behavior.
• Existing models (whole-brain, region-wise) have limitations in generalizability and interpretability for functional connectivity (FC) data.
• High dimensionality of FC data challenges whole-brain models, while region-wise models limit capturing integrated brain patterns.

Purpose of the Study:

• To develop an interpretable predictive model for fine-grained functional connectivity (FC) patterns associated with behavioral traits.
• To jointly learn regional relevance and prediction functions for a comprehensive characterization of FC-trait associations.
• To capture the collective contribution of brain-wide FC patterns to predicting cognitive traits.

Main Methods:

• Introduced a novel interpretable predictive model integrating regional and participant-level FC data.
• The model learns relevance scores and prediction functions for each brain region.
• Regional predictions are weighted by relevance scores for a participant-level prediction of cognitive traits.

Main Results:

• Validated the model using fMRI data from 6,798 participants in the Adolescent Brain and Cognitive Development (ABCD) study.
• Identified specific brain networks (cingulo-parietal, retrosplenial-temporal, dorsal attention, cingulo-opercular) as predictive of cognitive traits.
• Achieved competitive prediction accuracy and detailed characterization of fine-grained FC differences across cognitive domains.

Conclusions:

• The developed method effectively characterizes generalizable and fine-grained FC patterns linked to cognition in youth.
• Learned relevance scores improved predictions of longitudinal cognitive measures and traits in independent cohorts.
• The model offers enhanced interpretability and generalizability for understanding brain-behavior relationships.