GenCPM: a toolbox for generalized connectome-based predictive modeling.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Biostatistics

Background:

• Predicting cognitive and clinical outcomes from brain connectivity is crucial.
• Connectome-based Predictive Modeling (CPM) is widely used but limited to continuous outcomes and often ignores covariates.
• Existing CPM methods face challenges in clinical and disease cohort settings.

Purpose of the Study:

• Introduce GenCPM, a generalized Connectome-based Predictive Modeling framework.
• Extend CPM to support binary, categorical, and time-to-event outcomes.
• Integrate non-imaging covariates (demographic, genetic) to enhance prediction accuracy and interpretability.

Main Methods:

• Developed GenCPM using open-source R software.
• Incorporated marginal screening and regularized regression (LASSO, ridge, elastic net) for high-dimensional data.
• Applied GenCPM to Anti-Amyloid Treatment in Asymptomatic Alzheimer's Disease (A4) and Alzheimer's Disease Neuroimaging Initiative (ADNI) datasets.

Main Results:

• GenCPM demonstrated enhanced predictive performance compared to standard CPM methods.
• Improved signal attribution and interpretability were observed.
• The framework successfully handled diverse outcome types and integrated covariates.

Conclusions:

• GenCPM provides a flexible, scalable, and interpretable solution for predictive modeling in brain connectivity research.
• The generalized framework broadens the applicability of CPM in cognitive and clinical neuroscience.
• GenCPM facilitates more robust predictions of outcomes from neuromarkers.