Cross-cohort replicability and generalizability of connectivity-based psychometric prediction patterns

Area of Science:

• Neuroscience
• Cognitive Science
• Data Science

Background:

• Large-scale neuroimaging studies increasingly explore brain connectivity-behavior links.
• Replicability of identified brain-behavior associations is a critical, yet unresolved, issue.

Purpose of the Study:

• To assess cross-dataset replicability and generalizability of brain-behavior association patterns.
• To compare a region-wise approach with a whole-brain approach for prediction.
• To investigate predictions for fluid cognition and openness using neuroimaging data.

Main Methods:

• Utilized region-wise and whole-brain approaches for analyzing brain connectivity.
• Examined prediction models across multiple large population neuroimaging datasets.
• Assessed model generalizability through cross-dataset training and testing.

Main Results:

• Moderate similarity in fluid cognition prediction patterns across specific cohorts (HCP-YA, HCP-Aging, ENIGMA).
• Low similarity observed for openness prediction patterns across datasets.
• Moderate generalizability for fluid cognition predictions using the region-wise approach.
• Identified common brain regions associated with fluid cognition across cohorts.

Conclusions:

• Replicability and generalizability of connectivity-based predictions are moderate at best.
• Success is context-dependent, with fluid cognition showing more promise than openness.
• Further research is needed to improve the reliability and generalizability of these predictive models.