Introduction: Gamma-aminobutyric acid (GABA) is an essential inhibitory neurotransmitter that regulates neural communication and reduces neuron activity, reducing stress and promoting sleep. Two main types of inhibitory interneurons, Parvalbumin (PV+ ) and Somatostatin (SST+ ), shape brain responses to visual stimuli, although their specific roles remain unclear. Gathering quantitative data on their connectivity with pyramidal cells is vital for deciphering complex brain circuits. Understanding these neurons aids in developing treatments for neurological disorders and advancing intelligent machine design.
Methods: The study utilized data from the Allen Institute, employing the patch clamp technique on mice to analyze the firing characteristics of central inhibitory neurons, PV+  and SST+  Mice aged 45-70 days were anesthetized, perfused, and brain slices prepared. Researchers enhanced generalized leaky integrate-and-fire (GLIF) models to simulate neuronal spiking behavior, optimizing them with electrophysiological data to accurately reproduce neural responses across various stimuli.
Results: The analysis of explained variance indicates that additional mechanisms are needed to replicate the spiking behavior of inhibitory and excitatory neurons. The GLIF₁ model achieved a variance of 65%, with inhibitory neurons showing higher performance (82%) compared to excitatory neurons (69%). The GLIF₂ model’s reset rule reduced performance for both types, while the GLIF₃ model, including post-spike currents, improved inhibitory neuron variance to 84%. Introducing post-spike currents in the GLIF₄ model further enhanced performance for both neuron types. Overall, the findings highlight the distinct effects of post-spike currents on neuron types and the complexity of modeling their behaviors.
Conclusion: The present research concluded that GLIF models can accurately replicate biological neuron spike times with few adjustable parameters, simplifying input-output matching. More complex models, like GLIF₃, outperform simpler ones in differentiating transgenic lines. However, increasing complexity risks overfitting, complicating optimization. Clustering algorithms can classify cell types based on electrophysiological features, but identifying crucial features is essential for effective modeling.