Pretrained Models and Feature Datasets for ncpi

Description This dataset contains derived electrophysiological features, trained machine-learning inverse models, test-set predictions, and evaluation metrics generated from large-scale neural circuit simulations. Important note: The included features and machine-learning models have not yet been experimentally tested or independently validated. They are provided as research resources for reproducibility, benchmarking, and further methodological development. The resources were produced using the ncpi Python package (https://github.com/necolab-ugr/ncpi) to support neural circuit parameter inference from simulated electrophysiological signals. The original raw simulation signals are not included in this deposit. Instead, this record contains derived feature matrices and reusable trained inference models. Simulation datasets Results are provided for three simulation datasets: Hagen_v1Current-dipole-moment signals generated using the first version of the Hagen neural circuit model. This model uses current-based synapses. The Hagen_v1 dataset comprises approximately 2 million simulations generated by sampling neural circuit parameters within biologically plausible ranges. These ranges were constrained a priori according to biological principles. Model description: https://doi.org/10.1371/journal.pcbi.1010353 Hagen_v2Current-dipole-moment signals generated using the second version of the Hagen neural circuit model. As in Hagen_v1, this model uses current-based synapses. The Hagen_v2 dataset is a more constrained simulation set, comprising approximately 100,000 simulations. In addition to tighter parameter-range constraints, simulations were further filtered post hoc based on firing rates, spike-based metrics, and power-spectrum properties. Model description: https://doi.org/10.1371/journal.pcbi.1010353 CavallariProxy electrophysiological signals generated using the Cavallari neural circuit model. In contrast to the Hagen datasets, the Cavallari model uses conductance-based synapses. Model description: https://doi.org/10.3389/fncir.2014.00012 All signals were processed assuming a sampling frequency of 1600 Hz. Inferred neural circuit parameters Separate inverse models were trained for the Hagen and Cavallari simulation datasets. Although both model families estimate four common target variables — E_I, tau_exc, tau_inh, and external input strength — the underlying simulated parameters differ between the two neural circuit models. Hagen model Each Hagen simulation directly varies seven neural circuit parameters: J_EE: Strength of recurrent excitatory-to-excitatory connections.Range: 0.5–4.0 J_IE: Strength of excitatory-to-inhibitory connections.Range: 0.5–4.0 J_EI: Strength of inhibitory-to-excitatory connections.Range: −40.0 to −1.0 J_II: Strength of recurrent inhibitory-to-inhibitory connections.Range: −40.0 to −1.0 tau_syn_E: Excitatory synaptic time constant applied to excitatory inputs received by both neuronal populations.Range: 0.1–2.0 ms tau_syn_I: Inhibitory synaptic time constant applied to inhibitory inputs received by both neuronal populations.Range: 0.1–8.0 ms J_ext: Synaptic weight of the external Poisson input received by the network.Range: 10.0–50.0 The four recurrent coupling strengths are combined into the inferred excitatory-to-inhibitory target: E_I = (J_EE / J_EI) / (J_IE / J_II) Therefore, each Hagen inverse model estimates: E_I tau_exc, corresponding to tau_syn_E tau_inh, corresponding to tau_syn_I J_ext The individual recurrent coupling strengths are varied during simulation but are not estimated separately by the inverse models. Cavallari model Each Cavallari simulation modifies seven parameters relative to the default Cavallari configuration available in the ncpi repository: g_EE: Directly replaces exc_exc_recurrent, the excitatory-to-excitatory recurrent synaptic conductance.Values are sampled between 0.5 and 2.0 times the default value of 0.178. g_IE: Directly replaces exc_inh_recurrent, the excitatory-to-inhibitory recurrent synaptic conductance.Values are sampled between 0.5 and 2.0 times the default value of 0.233. g_EI: Directly replaces inh_exc_recurrent, the inhibitory-to-excitatory recurrent synaptic conductance.Values are sampled between 0.5 and 2.0 times the default value of −2.01. g_II: Directly replaces inh_inh_recurrent, the inhibitory-to-inhibitory recurrent synaptic conductance.Values are sampled between 0.5 and 2.0 times the default value of −2.70. tau_syn_AMPA_scale: Dimensionless scaling factor applied to tau_decay_AMPA in both excitatory and inhibitory neurons.Range: 0.5–2.0. The default AMPA decay constants are 2.0 ms for excitatory neurons and 1.0 ms for inhibitory neurons. tau_syn_GABA_scale: Dimensionless scaling factor applied to tau_decay_GABA_A in both excitatory and inhibitory neurons.Range: 0.5–2.0. The default GABA-A decay constant is 5.0 ms for both populations. ext_input_scale: Dimensionless scaling factor applied to both thalamic external-input conductances, th_exc_external and th_inh_external.Range: 0.5–4.0. The default values of these conductances are 0.234 and 0.317, respectively. Cortico-cortical external inputs are not modified. The four recurrent conductances are combined into the inferred excitatory-to-inhibitory target: E_I = (g_EE / g_EI) / (g_IE / g_II) Therefore, each Cavallari inverse model estimates: E_I tau_exc, corresponding to tau_syn_AMPA_scale tau_inh, corresponding to tau_syn_GABA_scale J_ext_equiv, corresponding to ext_input_scale The individual recurrent conductances are varied during simulation but are not estimated separately by the inverse models. Feature configurations Three feature configurations are included: catch22_22: Complete set of 22 canonical catch22 time-series features. specparam_3: Three spectral-parameterization features: aperiodic slope dominant peak frequency dominant peak power catch22_specparam_25: Concatenation of the 22 catch22 features and the 3 specparam-derived features. Spectral parameterization was performed over 5–200 Hz, with a goodness-of-fit threshold of R² ≥ 0.9. Invalid or non-finite feature rows were excluded before model training. Trained inverse models Four inverse-model types were trained for each dataset and feature configuration: Ridge regression MLPRegressor RandomForestRegressor Neural Posterior Estimation, NPE This produces up to 36 trained dataset–feature–model combinations: 3 datasets × 3 feature configurations × 4 model types Models were trained using a reproducible random 85% training / 15% testing split with random seed 0. The principal model configurations were: MLPRegressor: two hidden layers with 128 units each, ReLU activation, Adam optimizer, early stopping, and a maximum of 300 iterations. RandomForestRegressor: 300 trees, maximum depth of 24, and minimum leaf size of 2. Ridge: regularization parameter alpha = 1.0. NPE: neural spline flow estimator with five transforms and 50 hidden features, trained for up to 50 epochs. Included files The archive contains two main directories: simulation_datasets/new_features/ ML_models/new_models/ The feature directory includes: Computed feature matrices in NumPy format. Valid-row masks. Reproducible training and testing indices. Precomputed per-batch Hagen_v1 features. The model directory includes: Serialized trained models and feature scalers. NPE inference, density-estimator, and posterior assets. Model configuration files. Test target arrays, y_test.npy. Model predictions, y_pred.npy. Per-parameter and mean squared error metrics, metrics.json. Dataset-level result summaries, summary.json. Software The processing and training workflow was implemented using ncpi v1 beta. The ncpi source code is available at: https://github.com/necolab-ugr/ncpi The script to generate these results is also included.