NCPNet 1.0.2

Learning synapse-level brain circuit networks. Include training, inferring, evaluation, and visualization.

NCPNet

1. Brief Introduction

Neuronal Circuit Prediction Network (NCPNet), a simple and effective model for inferring neuron-level connections in a brain circuit network.

2. Environment and Dependencies

Our main dependencies:

torch==1.8.0
torch_geometric==2.0.1
torch-cluster==1.5.9
torch-sparse==0.6.12
torch-scatter==2.0.8
navis==1.3.1
neuprint-python==0.4.25

If you would like to reproduce our experiments and plots, please also install jupyter.

pip install jupyter

Code structure:

Source Code
├── data
|   ├──Hemibrain
|   └──C.Elegans
├── example
├── runs
├── configs
├── NCPNet
|   ├── approaches
|   ├── brain_data.py
|   ├── task.py
|   ├── trainer.py
|   └── utils.py
└── requirements.txt

Examples

NCPNet uses configuration files (yaml) to control training and test.

Run

python src/main_run.py -c src/configs/fly_linkpred.yaml

Reproducibility of Our Paper

Please try to use jupyter to reproduce our experiments in ./Plot_figure/

Access Data

Raw Data

The Drosophila connectome is available at https://www.janelia.org/project-team/flyem/hemibrain.

The C.elegans connectome is available at https://wormwiring.org/

Preprocessed Data

The data will be released after the review process.