Probabilistic mouse-human brain correspondence by multimodal optimal transport

Abstract

The mouse is the principal model for brain mechanism and disease, allowing experiments that cannot be performed in humans. However, findings often translate poorly because homologous regions differ in relative size and some human territories have no clear mouse counterpart. Here we present OTTER, which learns mouse-human brain correspondence as a probabilistic, parcel-resolution coupling using multimodal fused Gromov-Wasserstein optimal transport, integrating functional and structural connectivity with spatial position and curated homologies. OTTER recovers established homologues on a transcriptomic benchmark and preserves broad cross-species organisation along the cortical areal hierarchy. Applying the coupling to human functional connectivity reveals a graded decline in mouse-based reconstruction across evolutionarily expanded association cortex, with the lowest values in lateral prefrontal territory. Finally, the bidirectional map generates testable human predictions from mouse experiments and ranks mouse circuits corresponding to human clinical targets.