Unsupervised neuronal cell typing from connectivity by minimum description length

Connectomes are being reconstructed at rapidly increasing scale, creating a growing need for automated methods to identify neuronal cell types. Existing approaches rely heavily on expert annotation or on reference connectomes, and reliable unsupervised typing remains an open problem. We present a fully unsupervised method that recovers neuronal types using only connectivity, achieving 93-98% agreement with expert annotations on fruit fly optic lobes and visual systems. This is the first result to show that connectivity alone can achieve near perfect recovery of neuronal types in large regions of the brain. Our method is based on the minimum description length (MDL) principle: a good partition of neurons into clusters corresponds to a compression of the connectome. This provides a principled theoretical foundation for neuronal typing through the lens of compression. It also opens a path toward unsupervised typing in substantially harder regions, including the central brain and ventral nerve cord.