Can structure predict function at individual level in the human connectome?

Area of Science:

• Neuroscience
• Computational Neuroscience
• Brain Imaging

Background:

• Recent studies aim to predict individual brain Functional Connectivity (FC) from Structural Connectivity (SC).
• These predictive models promise enhanced performance and utility for individual-level brain mapping.
• The validity of these individual-level SC-to-FC mapping predictions requires rigorous investigation.

Purpose of the Study:

• To critically evaluate whether existing predictive models truly achieve meaningful individual-level mapping from SC to FC.
• To compare the performance of three prominent SC-to-FC predictive models against a simple group average predictor.
• To assess the robustness of additional analyses supporting the utility of the predictive models.

Main Methods:

• Utilized Human Connectome Project data shared across three relevant studies.
• Constructed a baseline predictor by averaging Functional Connectivity (FC) from training data.
• Analyzed the performance of the three selected predictive models and the group average predictor on individual-level FC prediction.

Main Results:

• The group average FC predictor demonstrated equivalent or superior raw prediction performance compared to the investigated individual-level models.
• Additional analyses by the original study authors, intended to show value beyond raw prediction, were reproducible using the group average predictor.
• The findings raise questions about the meaningfulness of the individual-level SC-to-FC mapping achieved by the evaluated methods.

Conclusions:

• Current methods may be inappropriate for the data, suffer from small sample sizes, or lack sufficient information for robust SC-to-FC mapping.
• Future individual-level studies must report performance against group averages and demonstrate genuine individual-level predictive information.
• Exploring alternative constructions for SC and FC is recommended to improve individual-level mapping.