Constructing mesoscale functionomics by neural dynamics subspace clustering

Building a fine-grained functional atlas of the brain is crucial for deciphering bio-intelligence. However, the intricate and non-linear interactions at single-neuron level lead to asynchronous and heterogeneous firing patterns among closely interacting neural populations, thereby challenging correlated-firing-based clustering methods. Here, we present Functional subspace clustering based on sparse Representation of Intrinsic Dynamics (FRID), an unsupervised approach for identifying neurons with shared microcircuit connectivity and information encoding properties (referred to as 'Functionomics') from mesoscale neural recordings. FRID significantly outperforms correlated-firing-based methods in both simulated complex networks and empirical calcium recordings. The accuracy and utility of FRID are validated across sensory integration and decision-making tasks to produce functionomic clusters with improved resolution and coding specificity than those defined by anatomical functional areas. As representative findings, we demonstrate single-cell level functional remapping during recovery from ischemic stroke and the learning-induced reorganization of functional modularization. Together with large-scale neural recordings, FRID could serve as a general data-driven paradigm to accelerate the understanding of mesoscale brain functions.