Combining multiple connectomes improves predictive modeling of phenotypic measures.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Biostatistics

Background:

• Functional connectivity matrices (connectomes) predict individual differences.
• Current methods often use only a single connectome, ignoring valuable data from multiple tasks.
• Combining information from different connectomes can enhance predictive power.

Purpose of the Study:

• To propose and validate a novel framework for integrating multiple task-based connectomes into a single predictive model.
• To enhance the accuracy of predicting phenotypic measures using brain connectivity data.
• To compare the proposed framework against existing connectome-based predictive modeling (CPM) approaches.

Main Methods:

• Developed a "multidimensional connectome-based predictive modeling" framework with two novel algorithms.
• Validated the framework using large datasets (Human Connectome Project, Philadelphia Neurodevelopmental Cohort) with multiple tasks.
• Compared performance against single-task CPM, averaged-connectome CPM, and a naïve multi-connectome extension.

Main Results:

• The proposed multidimensional framework demonstrated superior prediction performance compared to all competing methods.
• Different tasks contributed differentially to the predictive models, highlighting the importance of task selection.
• The study confirmed that integrating multiple connectomes yields better predictive accuracy.

Conclusions:

• The multidimensional connectome-based predictive modeling framework effectively combines information from multiple task connectomes.
• This approach offers a significant improvement over single connectome-based predictive models.
• Task selection is a critical factor in building accurate predictive models of individual differences.