Modeling Brain Microarchitecture with Deep Representation Learning

Models of neural architecture and organization are critical for many tasks in neuroscience. However, building these models in an automated, datadriven manner within and across varied brain regions still remains a challenge. In this work, we leverage the power of deep learning to build a rich model of neural microarchitecture across multiple, diversified brain areas. We then use low-rank matrix factorization to project the model’s features onto an interpretable, lower-dimensional space. Our results show that the subsequent embeddings possess biologically meaningful structure which makes them useful in the study of brain structure at multiple scales. We demonstrate the use of this approach in the discovery of microstructural patterns and motifs within brain areas, as well as in revealing relationships between multiple heterogeneous brain regions.