Reservoir computing model maps neuronal networks from neural activity

A cross-disciplinary team led by Ilya Auslender has introduced a new analytical method that fuses reservoir computing with graph theory to map functional connectivity in neuronal cultures. The approach, detailed in a preprint on arXiv (2608.09773), uses a Reservoir Computing (RC) framework originally proposed by the group in 2025 to extract an Intrinsic Connectivity Map (ICM) directly from multichannel electrophysiological recordings. The ICM acts as an effective adjacency matrix, representing how neurons influence one another. By applying centrality measures—such as node and edge importance metrics—the researchers quantified how individual neurons contribute to the culture's collective dynamics, linking these graph features to experimentally observed firing rates and other network-level activity descriptors.

To validate the inference, the team simulated the experimental environment, generating synthetic recordings from a known ground-truth connectivity matrix. This allowed them to benchmark the RC-derived connectivity against reality and assess how model performance changes with different graph structures. The results showed statistically robust, though varying-strength, associations between graph metrics and measured activity, supporting the validity of the RC-based approach. This scalable pipeline offers a powerful, data-driven framework for studying functional network properties in vitro, potentially advancing our understanding of how neuronal ensembles encode and process information. For AI researchers, the work also demonstrates how reservoir computing can be applied beyond traditional machine learning tasks to infer latent structure in complex biological systems.