Simulation data and machine learning models from 'A Python toolbox for neural circuit parameter inference'

Simulation datasets generated using the LIF network model, along with the trained machine learning models, supporting the article 'A Python Toolbox for Neural Circuit Parameter Inference'. Accepted for publication in npj Systems Biology and Applications.

The data folder contains features extracted from simulation data using two methods: catch22 and 1/f slope (stored in the power_spectrum_parameterization_1 folder). Each method-specific folder includes feature files labeled as sim_X and their corresponding parameter files labeled as sim_theta. Additionally, features extracted from EEG electrodes are organized in EEG subfolders within each method's directory.

The ML_models folder contains trained models, scalers, and—when using SBI approaches—density estimators, for various parameter inference configurations:

• 1_param/: models estimating a single parameter (E/I ratio)

• 2_param/: models estimating two parameters (E/I ratio and J_ext)

• 4_param/: models estimating four parameters (E/I ratio, time constants, and J_ext)

The EEG/ folder contains models trained specifically on features extracted from simulated EEG electrode data.

The held_out_data_models/ folder includes models trained on 90% of the simulation data, with the remaining 10% reserved for evaluation. The datasets/ subfolder within it contains the IDs of the held-out samples used for testing.

The code for loading and analyzing these data is available in the following repository: https://github.com/necolab-ugr/ncpi.