Model-based design of subthalamic nucleus neurons using hybrid optimization

The subthalamic nucleus (STN) is a key node of the basal ganglia and an established surgical target for the treatment of wide-ranging neurological and psychiatric disorders. However, scientific questions remain on the basis of STN action potential firing, its modulation by broader network activity patterns (e.g., cortical excitation vs. pallidal inhibition), and the generation of electrophysiological disease-state biomarker signals (e.g., elevated beta-band activity in Parkinson's disease). Computational models represent promising tools for exploring those kinds of questions, but the currently available STN neuron models have limitations in reproducing known electrophysiological properties. Therefore, we used a novel hybrid optimization framework, combining a genetic algorithm (GA) with simulation-based inference (SBI), to parameterize a four-compartment neuron model (soma, proximal dendrite, distal dendrite, and axon initial segment) that closely matches STN firing characteristics. The GA allowed us to efficiently explore a vast parameter space, while SBI characterized the underlying probabilistic distributions, offering insights into how individual parameters influence model behavior. We then used the probabilistic distributions to better understand how specific biophysical features of STN neurons translate into their unique firing characteristics. This new STN neuron model also represents an approachable tool for exploring the impact of cellular details (e.g., distal vs. proximal synaptic inputs) on larger-scale network activity patterns (e.g., bursts of beta activity in the STN local field potential).NEW & NOTEWORTHY The optimized STN neuron model reproduces key features observed in experimental recordings, including spontaneous firing, action potential shape, hyperpolarization response, frequency-current relationship, and phase response to simulated synaptic input. In addition, the hybrid optimization strategy generated probability parameter distributions that not only improved the overall optimization results but also provided critical insights into how biophysical parameters influence firing characteristics.