Sparse, Predictive, and Interpretable Functional Connectomics with UoI_Lasso.

Area of Science:

• Systems Neuroscience
• Computational Neuroscience
• Neuroimaging Analysis

Background:

• Functional brain networks derived from neural activity are crucial for understanding neurobiology.
• Current statistical methods for network inference often suffer from inaccurate feature identification and biased parameter estimates.
• The precise selection and estimation of connections (edges) between neural elements (nodes) are critical for valid network analysis.

Purpose of the Study:

• To introduce a novel framework, Union of Intersections (UoI), for robust statistical feature selection and estimation in network neuroscience.
• To enhance the accuracy and reduce bias in identifying functional neural connections.
• To develop a scalable and modular approach applicable to diverse neurobiological datasets.

Main Methods:

• The Union of Intersections (UoI) framework utilizes intersection and union operations for feature selection and estimation, respectively.
• UoI was implemented in the context of linear regression, termed UoI_Lasso.
• Extensive numerical investigations on synthetic data were performed to validate the method's performance.

Main Results:

• UoI_Lasso demonstrated tight control over false-positives and false-negatives in feature selection.
• Parameter estimates exhibited low bias and low variance, maintaining high prediction accuracy.
• Application to human electrocorticography and nonhuman primate single-unit recordings yielded sparse, predictive, and interpretable functional networks.

Conclusions:

• The Union of Intersections (UoI) framework provides a flexible, modular, and scalable solution for network inference in systems neuroscience.
• UoI_Lasso effectively generates interpretable and predictive functional connectivity networks from complex neural data.
• This approach advances the ability to accurately map brain structure and computation from neural activity.