Interpreting computational mechanism from multi-area neural recordings
Distilling computational mechanisms from neural recordings requires disambiguating intrinsic dynamics from external inputs. Here, we introduce Dynamically Interacting Systems (DynISys), a multi-compartment dynamical-systems model that infers per-area dynamics and structured inter-area communication from multi-area spiking activity. DynISys combines neural ordinary differential equation (ODE)…
Researchers have developed a new method called Dynamically Interacting Systems (DynISys) to interpret computational mechanisms from multi-area neural recordings. This model helps separate intrinsic dynamics from external inputs in neural activity. DynISys utilizes neural ordinary differential equation (ODE) latent dynamics and injective latent-to-neural mappings, enabling it to make accurate and reliable dynamical inferences.
When applied to simultaneous recordings from the dorsal premotor (PMd) and primary motor (M1) areas during a continuous reaching task, DynISys successfully reconstructs neuronal responses without introducing any spurious latent features. By analyzing the model's dynamics, the researchers were able to uncover potential mechanisms underlying motor planning and execution.
Specifically, PMd-to-M1 inputs position a stable fixed point in M1, which encodes the execution dynamics for the upcoming movement, and this is subsequently evoked by external inputs that are not part of PMd.
Furthermore, DynISys demonstrates consistent latents when fitted to neural recordings spanning multiple sessions and days. In conclusion, DynISys provides an interpretable account of distributed neural computation by distilling multi-area recordings into a mechanistic understanding of the underlying neural processes.
Written by urgent.news from bioRxiv's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
