Brain connectome alignment drops the shared-coordinate assumption: Ossa learns a rotation and a sparse node set, beating size-matched baselines on held-out ABCD alignment

The correspondence between structural and functional connectivity becomes a solvable subgraph-localization problem: within each functional community, find a small group of regions whose induced structural subgraph best matches the rotated functional connectivity. Earlier practice compressed structure-function coupling into one whole-connectome correlation or reported region-wise scalar scores; the former erases spatial variation, and the latter describes variation without directly naming a compact node set. In simulation with fully mismatched functional and structural coordinates, Ossa reached a node-set Jaccard of 0.861 and exact recovery in 0.40 of 10 replicates, while fixed-coordinate Ossa-noW reached 0.205 with exact recovery of 0.00.

The method estimates, inside each community derived from functional connectivity, an orthogonal transformation and a sparse node-selection vector at once; the transformation absorbs the coordinate mismatch between modalities and the vector delimits the structural subgraph. This extends the classical orthogonal Procrustes alignment to a setting where the alignment target is itself unknown, being an induced subgraph selected from structural connectivity rather than a given subgraph. The binary selection is relaxed to continuous weights, solved by projected block-coordinate ascent, and thresholded into nodes; the ABCD baseline cohort after quality control was split 60/20/20 on the 68-region Desikan-Killiany atlas and evaluated across 10 random seeds.

The demonstrated scope is simulated data and the structural and functional connectomes of the ABCD baseline and Year-2 follow-up on the 68-region Desikan-Killiany atlas; the identities of the selected regions depend on the functional-community construction and the sparsity budget, and the authors state that replication across parcellations, acquisition pipelines and clinical cohorts is needed before assigning definitive biological significance.

The theoretical guarantees cover canonical Ossa within a fixed population community and concern only local maximizers in a specified neighborhood; the authors explicitly do not claim that the multi-restart block-ascent algorithm reaches that neighborhood or returns such a maximizer. The multi-clique extension serves as a diagnostic, and in ABCD the supports collapse into a single core at small penalties.