Identifiability of dendritic parameters from sparse intracellular measurements — and the numerical artifacts that inflate it
Year: 2026
Type: preprint
Abstract: Biophysical neuron models fitted to somatic recordings leave dendritic properties largely unconstrained. In a published all-active model from the Allen Cell Types Database, the apical tuft is electrically disconnected from the soma: steady-state voltage attenuation from soma to distal tuft is 3.2 × 10⁻⁷. We ask which intracellular dendritic measurements, and what voltage precision, are needed to identify the parameters that govern a neuron's response to dendritic input. Using a detailed layer 5b pyramidal-cell model (Hay et al., 2011) as ground truth and a reduced tree model validated against a reference converged in space, time step and initial state, we compute Fisher information for 30 region-level parameters, propagate it to a set of functional outputs, and test the linear approximation with the relative intrinsic curvature of Bates and Watts. With voltage recorded every 25 µm along the tree and current injected at the soma and at up to 15 dendritic sites, the specification is certified for voltage noise up to 8.6 mV when the noise is white, 1.9 mV when it is correlated over 1 ms, and 0.72 to 0.76 mV when it is correlated over 10 to 50 ms. Under slow noise the limit is curvature, not information: the Cramér–Rao requirement alone would tolerate about eight times more noise, and dendritic injection sites beyond the first five do not raise the certified level. Somatic recording alone fails the necessary condition in every design and noise model we examined. The requirement hinges on one property — the latency of the first spike of a dendritic calcium burst — whose spike count changes within the physiological range of the parameters; without it, the requirement relaxes at least 5- to 18-fold. We also document three numerical artifacts — an unconverged resting state, single-precision storage and integrator noise — each of which inflates apparent identifiability, and give a control for each.
Author: Del Colle, Martel Alexandre
Language: English
Subfield: Cognitive Neuroscience
Field: Neuroscience
Domain: Life Sciences
Open Access status: green
OpenAlex ID: W7234632178
