Toward a multisubject analysis of neural connectivity

Area of Science:

• Neuroscience
• Statistics
• Machine Learning

Background:

• Directed acyclic graphs (DAGs) and probability models are crucial for analyzing neural connectivity.
• Multisubject studies often involve related connectivities, necessitating methods to leverage shared features for statistical efficiency.

Purpose of the Study:

• To present and discuss the implications of an exact algorithm for estimating multiple related DAGs in multisubject settings.
• To apply and illustrate the methodology using functional magnetic resonance imaging (fMRI) data from a multisubject experiment.

Main Methods:

• Utilizing an exact algorithm for the estimation of multiple related DAGs, building upon prior work by Oates et al. (2014).
• Applying the methodology to multisubject fMRI data, including joint learning of subject-specific connectivity.
• Estimating relationships between subjects within heterogeneous collections.

Main Results:

• Demonstrated the application of the exact DAG estimation methodology to multisubject fMRI data.
• Illustrated methods for retrospective elicitation of tuning parameters using technical replicate data.
• Showcased the simultaneous estimation of subject-specific connectivities and inter-subject relationships.

Conclusions:

• The exact estimation of multiple related DAGs offers significant potential for analyzing complex multisubject neuroimaging data.
• The methodology allows for joint learning of individual connectivities and inter-subject relationships, enhancing statistical efficiency.
• Careful elicitation of tuning parameters is important for successful application in multisubject settings.