AutoMIND: model configurations, simulations, target observations (experimental data), and trained deep generative models

Data here include:Training (prior samples) and discovered (posterior samples) parameter configurations of the clustered-AdEx spiking neural network,summary features of the corresponding simulations,summary features of the target observations (on which posterior density estimators are conditioned),and trained deep generative models (Normalizing Flow-based neural density estimators).For details, see: Gao et al., 2024, Deep inverse modeling reveals dynamic-dependent invariances in neural circuit mechanisms.<br>Data directories are organized as follows:./training_prior_samples/training.zip: 1-million parameter configurations and summary features from model simulations used to train deep generative models.heldout.zip: additional network simulations not used for DGM training, a subset of which was used as synthetic observations../discovered_posterior_samples/organoids.zip: discovered model configurations consistent with human brain organoid network burst across development. See ./organoid_predictives.png.mouse-vis.zip: discovered model configurations consistent with population firing rate PSD of Neuropixels recordings from mouse visual areas. See ./allen_predicitves_all.png.mouse-hc.zip: discovered model configurations consistent with population firing rate PSD of Neuropixels recordings from mouse hippocampal areas. See ./allen_predicitves_all.png.synthetic.zip: discovered model configurations consistent with population firing rate PSD of synthetic observations, i.e., held-out network simulations. See ./synthetic.png.<b>NOTE</b>: all zip files also contain the prior distribution, posterior density estimator, and config files necessary for running the simulations and analyses../dgms/burst_posterior.pickle: trained conditional density estimator that approximates the posterior distribution conditioned on network burst summary features.psd_posterior.pickle: trained conditional density estimator that approximates the posterior distribution. conditioned on network firing rate power spectral densities../observations/allenvc_summary: population firing rate PSD of Neuropixels recordings.synthetic_summary: various summary features of synthetic observations (i.e., held-out network simulations).organoid_summary: population firing burst statistics of organoid multi-electrode array recordings.<br>It's recommended to keep the full path when downloading and unzipping for consistency with e.g., code demo.Note that the raw simulation (i.e., spike times from recorded neurons) data files are too large and exceed the figshare upload file size limit. They will be provided elsewhere and linked here.