Skip to content
Branch: master
Find file History
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.

Graph Learning on Biological Networks


We train each model on one tesla V100.

For training the default ResMRConv-14 with 64 filters, run

python -u examples/ppi/ --phase train --data_dir /data/deepgcn/ppi

If you want to train model with other gcn layers (for example EdgeConv, 28 layers, 256 channels in the first layer, with dense connection), run

python -u examples/ppi/ --phase train --conv edge --data_dir /data/deepgcn/ppi  --block dense --n_filters 256 --n_blocks 28

Just need to set --data_dir into your data folder, dataset will be downloaded automatically. Other parameters for changing the architecture are:

--block         graph backbone block type {res, plain, dense}
--conv          graph conv layer {edge, mr, sage, gin, gcn, gat}
--n_filters     number of channels of deep features, default is 64
--n_blocks      number of basic blocks, default is 28


Pretrained Models

Our pretrained models can be found from Goolge Cloud.

The Naming format of our pretrained model: task-connection-conv_type-n_blocks-n_filters_phase_best.pth, eg. ppi-res-mr-28-256_val_best.pth, which means PPI node classification task, with residual connection, convolution is MRGCN, 28 layers, 256 channels, the best pretrained model found in validation dataset.

Use parameter --pretrained_model to set the specific pretrained model you want.

python -u examples/ppi/ --phase test --pretrained_model checkpoints/ppi-res-mr-28-256_val_best.pth --data_dir /data/deepgcn/ppi --n_filters 256 --n_blocks 28 --conv mr --block res
python -u examples/ppi/ --phase test --pretrained_model checkpoints/ppi-dense-mr-14-256_val_best.pth --data_dir /data/deepgcn/ppi --n_filters 256 --n_blocks 14 --conv mr --block dense

Please also specify the number of blocks and filters according to the name of pretrained models.

You can’t perform that action at this time.