Bayesian inference for biophysical neuron models enables stimulus optimization for retinal neuroprosthetics.

Area of Science:

• Computational neuroscience
• Retinal biophysics
• Neuroprosthetics

Background:

• Multicompartment neuron models are crucial for studying neuronal biophysics.
• Parameter inference for these models, especially with uncertainty, remains a significant challenge.
• Understanding retinal circuitry is key for developing effective neuroprosthetics.

Purpose of the Study:

• To apply Bayesian inference for parameter estimation of detailed mouse retinal neuron models.
• To obtain multivariate posterior distributions for model parameters, including uncertainty.
• To develop data-driven neuron models for optimizing retinal neuroprosthetics.

Main Methods:

• Bayesian inference was employed to estimate parameters for photoreceptor, OFF-cone bipolar cell, and ON-cone bipolar cell models.
• Two-photon imaging data from the mouse retina was used for model fitting.
• A simulation environment for external electrical stimulation was created.

Main Results:

• Multivariate posterior distributions for model parameters were successfully obtained.
• Plausible parameter ranges consistent with the data were identified.
• Parameters poorly constrained by the data were highlighted.
• Optimized stimulus waveforms were generated to selectively target OFF- and ON-cone bipolar cells.

Conclusions:

• Bayesian inference provides a robust framework for parameterizing detailed neuron models with uncertainty.
• Mechanistic, data-driven neuron models are valuable for advancing retinal neuroprosthetics.
• The developed models and simulation environment can guide the design of targeted retinal stimulation strategies.