Interpreting computational mechanism from multi-area neural recordings

DynISys infers per-area neural dynamics and structured PMd-to-M1 communication from simultaneous spiking during continuous reaching, reconstructing responses without inventing spurious latent features.

• Why it matters: Neural recordings mix intrinsic dynamics with external inputs, making computational mechanism inference ambiguous and potentially misleading.
• What they did: The method uses a multi-compartment dynamical-systems model with neural ODE latent dynamics and injective latent-to-neural mappings, then fits it to PMd and M1 spiking from a continuous reaching task.
• The result: Model dynamics suggest PMd-to-M1 inputs position a stable fixed point in M1 that encodes execution dynamics for the upcoming movement, and multi-session fits yield consistent latents across multiple days.

AI summary of the abstract · may contain errors · not medical advice · read the paper.