TransBrain: A computational framework for translating brain-wide phenotypes between humans and mice

Abstract

Despite remarkable advances in whole-brain imaging technologies, the lack of quantitative approaches to bridge rodent preclinical and human studies remains a critical challenge. Here we present TransBrain, a computational framework enabling bidirectional translation of brain-wide phenotypes between humans and mice. TransBrain improves human-mouse homology mapping accuracy through: (1) a novel detached region-specific deep neural network trained on integrated multi-modal human transcriptomics to improve cortical correspondence (89.5% improvement over the original transcriptome), which revealed two evolutionarily conserved gradients explaining >50% of cortical organizational variance, and (2) random walk-based graph representation learning to construct a unified cross-species latent space incorporating anatomical hierarchies and structural connectivity. We demonstrated TransBrains utility through three cross-species applications: quantitative assessment of resting-state brain organizational features, inferring human cognitive functions from mouse optogenetic circuits, and translating molecular insights from mouse models to individual-level mechanisms in autism. TransBrain enables quantitative cross-species comparison and mechanistic investigation of both normal and pathological brain functions.