-
Notifications
You must be signed in to change notification settings - Fork 0
Inferring Monosynaptic Connectivity from Cross correlograms
We use a MAT neuron network with 1000 neurons as the model hypothesis and simulate the network for 2h at the time step of 0.025 ms (Endo et al., 2021). The original code of the network model should be addressed to the correspondence of Endo et al. (2021).
Different from biophysical property inference, the simulation of the training data is independent from code of our inference framework. We put the sample training data used in our paper (Sheng et al., 2025) at the directory: data/syn_data. The cross-correlograms (CCGs) are stored under the key 'dynamic_data' and the labels are stored under the key 'target_data' which are binary. The data are accessed via PyTorch dataloader class SynDataset(Data.Dataset) in utils.py.
The experimental data used in (Sheng et al., 2025) are obtained from Buzsaki's lab (see link for details). We selected and put the experimental CCRs at the directory: data/exp_data. The construct_exp(exp_test_data_path, batch_size) function in main.py and the dataloader : will get access to the data. You can put your own experimental data here using the same format.
So far, we've already prepared everything to use Domain Adaptive Neural Inference Framework to infer monosynaptic connectivity. Now, we need to register the neural simulator we've constructed in main.py. Simply run the command line:
CUDA_VISIBLE_DEVICES={cuda_id} python main.py -t 100 -b 800 -seq 1 -e 100 -dir /you/save/path -lr 1e-4 -size 385020 -syn_data ./data/syn_data/data/MAT_1000_T_0.1.h5 -cnn_type resnet -cnn_config 18 -time_type lstm -out 1 -fl 0 -dynamic_key dynamic_data -target_key target_data -exp_test_data ./data/exp_data/Hip_CA1_pos_9_neg_66_20220303_spon.h5 -subdata 1 -da 1 -da_factor {da_factor} -da_type CAN -da_n_kernel 5 -da_kernel_mul 2 -da_thre {da_thre} -da_eps 1e-6 -da_n_layer 1 -da_layer_offset 0 -st 1 -st_factor 1.0 -st_thre {st_thre} -start_idx 0 -method aug -min_thre 0.01 -seed 0You can view the intermediate results via tensorboard or in your /you/save/path.
Endo, D., Kobayashi, R., Bartolo, R., Averbeck, B. B., Sugase-Miyamoto, Y., Hayashi, K., ... & Shinomoto, S. (2021). A convolutional neural network for estimating synaptic connectivity from spike trains. Scientific Reports, 11(1), 1-18.