Uncovering hidden functional brain organization by random matrix theory

In the brain, systems are organized in a modular way, serving different functionalities. We present a novel method, based on random matrix theory, for the detection of functional modules in neural systems. Our method can efficiently filter out both local unit-specific noise and global system- wide dependencies that typically obfuscate the presence of sub-structures, thus unveiling a hidden functional structure. Using correlation-based community detection, the method yields an optimally contrasted mesoscopic functional structure, i.e. groups of neurons (or small brain regions) that are positively correlated internally, while at the same time negatively correlated with each other. The method is purely data driven and does not use any presumptions or network representation. In an experimental dataset containing time series of neuronal gene expression, the method detects two hidden functional modules which are recognized as existing anatomical regions, without any prior anatomical knowledge(, showing the potential of the method to be used on brain data).