biophysical-hh-model-extracellular-neurostimulation

name	biophysical-hh-model-extracellular-neurostimulation
description	Learning biophysical Hodgkin-Huxley models from extracellular MEA measurements using differentiable simulation and simulation-based inference. Enables precise neurostimulation prediction from minutes of recording instead of hours of stimulus testing.
trigger_words	["biophysical HH model","extracellular MEA","neurostimulation prediction","differentiable biophysical simulation","Hodgkin-Huxley inference","simulation-based inference","multi-electrode array HH fitting","precise neurostimulation","neural dynamics prediction"]
categories	["neuroscience","computational neuroscience","neuroengineering"]
created	2026-07-12
source	arXiv:2607.04063v1 (ICML 2026)

Learning Biophysical Hodgkin-Huxley Models from Extracellular MEA Data

Paper

Title: Learning Biophysical Models of Large-Scale Multineuronal Data to Enable Precise Neurostimulation
Authors: Amrith Lotlikar, Ian Christopher Tanoh, Praful Vasireddy, Andrew Lanpouthakoun, Ramandeep Vilkhu, Michael Sommeling, A.J. Phillips, Alexander Sher, Alan Litke, Scott W. Linderman, E.J. Chichilnisky, Subhasish Mitra
Published: arXiv:2607.04063v1 (Accepted at ICML 2026)
Date: July 5, 2026

Problem

Multi-compartment Hodgkin-Huxley (HH) models provide a principled framework for predicting neural dynamics and responses to electrical stimulation. However:

• Invasive requirement: Fitting HH biophysical parameters traditionally requires intracellular recordings, which are invasive and low-throughput
• Scalability limit: Cannot capture the geometry and cell-specific properties of many neurons in a given neural circuit simultaneously
• Clinical inefficiency: Predicting neural spiking responses to candidate neurostimulation patterns takes hours of clinical testing

Solution

A framework to rapidly infer HH parameters from designed features of extracellular MEA measurements using:

1. Differentiable biophysical simulation — Enables gradient-based optimization of HH parameters through the full biophysical model
2. Simulation-based inference (SBI) — Bayesian inference framework that learns the posterior over biophysical parameters from extracellular data alone
3. Extracellular-only fitting — No intracellular recordings needed; uses only high-density extracellular measurements from full neural populations

Key Results

• 90.6% accuracy predicting previously unseen multi-electrode stimulation responses
• HH models fit from only a few minutes of recording (vs. hours of stimulus testing)
• Validated on hundreds of hours of stimulation and recording data from isolated macaque retina
• Used 30 μm-pitch 512-electrode array for high-density extracellular measurements

Methodology Details

Differentiable Biophysical Simulation

The key innovation is making the full multi-compartment HH model differentiable, enabling:

• Gradient-based parameter estimation — Compute gradients of the simulation output with respect to biophysical parameters (ion channel conductances, membrane capacitance, etc.)
• End-to-end optimization — Directly optimize HH parameters to match extracellular voltage traces
• Amortized inference — Once trained, the inference model can rapidly predict parameters for new neurons

Simulation-Based Inference (SBI)

SBI learns a mapping from observed data features to the posterior distribution over HH parameters:

• Feature extraction — Designed features from extracellular MEA measurements (spike waveforms, temporal patterns, spatial spread)
• Posterior learning — Neural density estimator trained on simulated data pairs (parameters → features)
• Amortized inference — Single inference model works across many neurons without re-fitting

Validation Pipeline

1. Data collection: Hundreds of hours of stimulation + recording from macaque retina (512-electrode array, 30 μm pitch)
2. Parameter fitting: Fit HH models from minutes of spontaneous + stimulated recording
3. Prediction validation: Predict responses to novel stimulation patterns never seen during fitting
4. Accuracy metric: 90.6% prediction accuracy on held-out stimulation responses

Practical Applications

Translational Neuroengineering

• Clinical neurostimulation planning: Predict patient-specific responses to candidate stimulation patterns before delivery
• Deep brain stimulation (DBS): Optimize DBS parameters using patient-specific biophysical models
• Retinal prosthetics: Predict response patterns for retinal implant stimulation

Research Applications

• Circuit-level modeling: Fit biophysical models for entire neural populations from extracellular data
• Cell-type classification: Extract biophysical signatures for cell-type identification
• Drug effect prediction: Model how pharmacological interventions alter HH parameters

Implementation Notes

• Differentiable simulators: Tools like JAX-based HH simulators or TorchDiffEq for gradient computation
• SBI libraries: sbi (simulation-based inference) Python package, or custom neural density estimators
• MEA data: Requires high-density extracellular recordings (e.g., Neuropixels, CMOS arrays)
• Compute: Training requires significant GPU resources for differentiable simulation + SBI

Trigger Conditions

Use this skill when:

• Inferring biophysical neuron model parameters from extracellular recordings
• Predicting neural responses to electrical stimulation patterns
• Designing neurostimulation protocols for clinical or research applications
• Building differentiable biophysical neuron simulators
• Applying simulation-based inference to neuroscience problems

References

• Paper: arXiv:2607.04063v1 (ICML 2026)
• Related: pinn-neuronal-parameter-estimation (PINN-based HH parameter estimation from partial voltage observations)
• Related: differentiable-biophysical-simulation-neurostimulation (Differentiable biophysical simulation framework)