Translating brain anatomy and neurodegenerative disease from mouse to human through latent gene expression space

The mouse model is by far the most widely used animal model in preclinical neuroscience, but translating findings to humans suffers from the lack of a formal framework establishing the correspondence between the mouse and the human brain. In this study, we build on the concept of common brain space, and on previous work embedding gene expression profiles, to bring the two species into alignment for comparative analysis. Using a variational autoencoder (VAE) combined with a latent classifier, we create a latent space that strikes a balance between abstract features related to reconstruction and features pertaining to regional segregation. This approach demonstrates improved cross-species homology and within-species locality compared to existing comparative models. In addition, we show that brain alterations in mouse disease models can be translated to humans, predicting patterns of brain changes in Alzheimer's and Parkinson's diseases. The flexibility and scalability of this approach offer a promising framework to bridge between more animal models, comparing quantitative imaging modalities, and disease phenotypes. This in turn will help advance our understanding of species similarities and differences, enhancing both fundamental translational neuroscience and disease modelling.