Neuronal activity across the brain during behavior can be readily accessed with electrophysiology or calcium imaging. However, tools to infer how various brain regions communicate with each other during behavior are lacking. To address this shortcoming, Kanaka Rajan at Harvard Medical School in Boston and her colleagues developed CURBD (current-based decomposition) in a multi-laboratory collaboration.
CURBD harnesses recurrent neural networks (RNNs) to model experimentally acquired neuronal activity. Rather than considering neuronal outputs (as do other neuronal data analysis tools), CURBD focuses on neuronal inputs. Communication between neurons is modeled as activations and interaction strengths or weights in an RNN.
To model brain-wide interactions, this RNN consists of a network of networks, which represent different brain regions. After training the RNN with calcium imaging or electrophysiology data recorded in different brain regions during behavior, the activity in one brain region can be inferred from the activity in other regions as the RNN has learned the connections between these regions. This is a preview of subscription content, access via your institution Access Nature and 54 other Nature Portfolio journals Get Nature+, our best-value online-access subscription Receive 12 print issues and online access Prices may be subject to local taxes which are calculated during checkout Nature Methods https://www.nature.com/nmeth/ Vogt, N.
Extracting brain-wide interactions with CURBD. Nat Methods 23, 1938 (2026). https://doi.org/10.1038/s41592-026-03258-9 DOI: https://doi.org/10.1038/s41592-026-03258-9
Extract — continue reading at the source.
