Joint embedding: A scalable alignment to compare individuals in a connectivity space.

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Brain Connectomics

Background:

• Understanding human brain organization requires a common coordinate space for comparing functional data across individuals.
• Dimensionality reduction of fMRI data can represent individual functional architecture but aligning these representations while preserving individual specificity is challenging.
• Resting-state fMRI (rs-fMRI) data presents unique challenges for cross-individual functional feature comparison.

Purpose of the Study:

• To develop and evaluate a scalable joint embedding method for aligning multiple individual brain connectomes into a common space.
• To assess if joint embedding preserves individual-specific connectivity structures while enabling cross-individual comparisons.
• To demonstrate the utility of the common space for analyzing functional changes across time, tasks, and lifespan.

Main Methods:

• Proposed a scalable joint embedding algorithm to simultaneously embed multiple individual brain connectomes.
• Utilized Human Connectome Project (HCP) data for evaluation and comparison with the orthonormal alignment model.
• Applied the method to rs-fMRI data and a lifespan cohort (ages 6-85) for age prediction analysis.

Main Results:

• Joint embedding significantly increased the similarity of functional representations across individuals compared to the orthonormal alignment model.
• The method successfully captured distinct individual profiles, enhancing discriminability between participants.
• The common space derived from rs-fMRI showed improved overlap of task-evoked activation across participants.
• Joint embedding achieved better age prediction (r² = 0.65) in a lifespan cohort than the prior alignment model, facilitating the characterization of functional trajectories.

Conclusions:

• Scalable joint embedding effectively aligns functional brain data across participants and populations within a common connectivity space.
• The approach simultaneously captures individual neural representations and preserves individual specificity.
• This method offers a powerful tool for analyzing individual differences, functional changes over time, and developmental trajectories in the human brain.