The Study of Drug-Drug Interaction Learning Through Various Graph Learning Methods

Setup:

0: Install cuda driver like CUDA Toolkit 10.2

Install instruction in Linux: 
wget https://developer.download.nvidia.com/compute/cuda/10.2/Prod/local_installers/cuda_10.2.89_440.33.01_linux.run
sudo sh cuda_10.2.89_440.33.01_linux.run

Update environment variable (you can add it into  ~/.bashrc): 
LD_LIBRARY_PATH="$LD_LIBRARY_PATH:/usr/local/cuda-10.2/lib64"
PATH="$PATH:/usr/local/cuda-10.2/bin"

1: Install from requirements.txt (using conda or pip install)

2: Install pytorch

On MAC or Linux, it's very simple:

pip install torch==1.6.0 

On Windows, please follow this link pytorch:
No CUDA To install PyTorch via pip, and do not have a CUDA-capable system or do not require CUDA, in the above selector, choose OS: Windows, Package: Pip and CUDA: None. Then, run the command that is presented to you.

With CUDA To install PyTorch via pip, and do have a CUDA-capable system, in the above selector, choose OS: Windows, Package: Pip and the CUDA version suited to your machine. Often, the latest CUDA version is better. Then, run the command that is presented to you.

3: Install pytorch-geometric

Ensure that at least PyTorch 1.4.0 is installed:

$ python -c "import torch; print(torch.__version__)"
>>> 1.6.0

Find the CUDA version PyTorch was installed with:

$ python -c "import torch; print(torch.version.cuda)"
>>> 10.2

Install the relevant packages:

pip install torch-scatter -f https://pytorch-geometric.com/whl/torch-${TORCH}+${CUDA}.html
pip install torch-sparse -f https://pytorch-geometric.com/whl/torch-${TORCH}+${CUDA}.html
pip install torch-cluster -f https://pytorch-geometric.com/whl/torch-${TORCH}+${CUDA}.html
pip install torch-spline-conv -f https://pytorch-geometric.com/whl/torch-${TORCH}+${CUDA}.html
pip install torch-geometric

where ${CUDA} and ${TORCH} should be replaced by your specific CUDA version (cpu, cu92, cu101, cu102, cu110) and PyTorch version (1.4.0, 1.5.0, 1.6.0, 1.7.0), respectively. For example, for PyTorch 1.7.0/1.7.1 and CUDA 11.0, type:

For example in my case, I have PyTorch 1.6.0 and CUDA 10.2, so I will install with following commands:

pip install torch-scatter -f https://pytorch-geometric.com/whl/torch-1.6.0+cu102.html
pip install torch-sparse -f https://pytorch-geometric.com/whl/torch-1.6.0+cu102.html
pip install torch-cluster -f https://pytorch-geometric.com/whl/torch-1.6.0+cu102.html
pip install torch-spline-conv -f https://pytorch-geometric.com/whl/torch-1.6.0+cu102.html
pip install torch-geometric

pytorch-geometric Reference

4: Install ogb

pip install ogb
python -c "import ogb; print(ogb.__version__)"
Note: This should print "1.2.6". Otherwise, please update the version by
pip install -U ogb

### Dataset ogbl-ddi:
  1. ```https://ogb.stanford.edu/docs/linkprop/#ogbl-ddi

Run commands with parameters (corresponding to the best runs only):

 1. python graphsage_augmented_node2vec.py --hidden_channels 512 --num_layers 2 --dropout 0.5 --lr 0.005 --batch_size 65536 --epoch 200 --runs 10
 2. python deepergcn_augmented.py --hidden_channels 64 --num_layers 2 --dropout 0.5 --lr 0.005 --batch_size 65536 --epoch 200 --runs 10 
 3. python gnn_augmented_node2vec_random.py --hidden_channels 512 --num_layers 2 --dropout 0.5 --lr 0.005 --batch_size 65536 --epoch 200 --runs 10 
 4. python gnn_augmented_node2vec_skip.py --hidden_channels 512 --num_layers 2 --dropout 0.55 --lr 0.005 --batch_size 65536 --epoch 200 --runs 10
 5. python gnn_augmented_node2vec.py --hidden_channels 1024 --num_layers 2 --dropout 0.5 --lr 0.005 --batch_size 65536 --epoch 200 --runs 10 

Reference performance for OGB:

Best Runs

Model (command)	Test Hits@20 (%)	Validation Hits@20(%)	Parameters	Hardware
MAD Learning (baseline)	67.81 ± 2.94	70.10 ± 0.82	1,228,897	Geforce GTX 1080 Ti (11GB GPU)
LRGA + GCN (baseline)	62.30 ± 9.12	66.75 ± 0.58	1,576,081	Tesla P100 (16GB GPU)
LRGA + GraphSage + Node2Vec (1)	61.23 ± 13.62	68.27 ± 0.96	Tesla V100 (32GB)
Deeper GCN Augment (2)	31.52 ± 8.27	56.92 ± 1.33	319,555	Tesla V100 (32GB)
LRGA + GCN Aug + Node2Vec + RNN (3)	70.66 ± 5.88	69.58 ± 0.90	3,807,121	Tesla V100 (32GB)
LRGA + GCN Aug + Node2Vec:SkipConnect (4)	65.91 ± 11.22	71.66 ± 1.38	3,126,985	Tesla V100 (32GB)
LRGA + GCN Aug + Node2Vec (5)	73.85 ± 8.71	72.25 ± 0.47	10,235,281	Tesla V100 (32GB)

All Runs

Model	Test Hits@20 (%)	Validation Hits@20(%)	Parameters	Hardware
MAD Learning (baseline)	67.81 ± 2.94	70.10 ± 0.82	1,228,897	Geforce GTX 1080 Ti (11GB GPU)
LRGA + GCN (baseline)	62.30 ± 9.12	66.75 ± 0.58	1,576,081	Tesla P100 (16GB GPU)
LRGA + GraphSage + Node2Vec	61.23 ± 13.62	68.27 ± 0.96	Tesla V100 (32GB)
Deeper GCN Augment	31.52 ± 8.27	56.92 ± 1.33	319,555	Tesla V100 (32GB)
LRGA + GCN Aug	55.08 ± 13.44	66.03 ± 2.48	1,576,081	Tesla V100 (32GB)
LRGA + GCN Aug	66.55 ± 8.70	69.85 ± 0.60	3,807,121	Tesla V100 (32GB)
LRGA + GCN Aug + Node2Vec + RNN	70.66 ± 5.88	69.58 ± 0.90	3,807,121	Tesla V100 (32GB)
LRGA + GCN Aug + Node2Vec:SkipConnect	65.33 ± 8.21	71.19 ± 0.57	3,126,985	Tesla V100 (32GB)
LRGA + GCN Aug + Node2Vec:SkipConnect	65.91 ± 11.22	71.66 ± 1.38	3,126,985	Tesla V100 (32GB)
LRGA + GCN Aug + Node2Vec:SkipConnect	69.61 ± 04.39	70.51 ± 1.00	3,126,985	Tesla V100 (32GB)
LRGA + GCN Aug + Node2Vec:SkipConnect	68.74 ± 6.47	70.46 ± 1.42	3,126,985	Tesla V100 (32GB)
LRGA + GCN Aug + Node2Vec:SkipConnect	23.34 ± 24.69	68.85 ± 0.65	4,749,913	Tesla V100 (32GB)
LRGA + GCN Aug + Node2Vec	56.22 ± 10.15	65.52 ± 0.87	1,576,081	Tesla V100 (32GB)
LRGA + GCN Aug + Node2Vec	60.80 ± 10.45	67.28 ± 0.74	3,807,121	Tesla V100 (32GB)
LRGA + GCN Aug + Node2Vec	73.41 ± 7.15	70.13 ± 0.50	3,807,121	Tesla V100 (32GB)
LRGA + GCN Aug + Node2Vec	65.91 ± 10.11	70.98 ±0.65	3,807,121	Tesla V100 (32GB)
LRGA + GCN Aug + Node2Vec	67.96 ± 10.41	71.33 ± 0.52	3,807,121	Tesla V100 (32GB)
LRGA + GCN Aug + Node2Vec	62.69 ± 5.65	67.28 ± 2.10	3,807,121	Tesla V100 (32GB)
LRGA + GCN Aug + Node2Vec	45.18 ± 14.67	62.29 ± 6.09	4,749,913	Tesla V100 (32GB)
LRGA + GCN Aug + Node2Vec	72.91 ± 5.07	70.62 ± 1.01	3,807,121	Tesla V100 (32GB)
LRGA + GCN Aug + Node2Vec	50.33 ± 13.37	67.52 ± 0.75	3,126,985	Tesla V100 (32GB)
LRGA + GCN Aug + Node2Vec	50.33 ± 13.37	67.52 ± 0.75	3,126,985	Tesla V100 (32GB)
LRGA + GCN Aug + Node2Vec	64.48 ± 14.18	65.32 ± 5.56	4,749,913	Tesla V100 (32GB)
LRGA + GCN Aug + Node2Vec	73.41 ± 7.15	70.13 ± 0.50	3,807,121	Tesla V100 (32GB)
LRGA + GCN Aug + Node2Vec	73.51 ± 8.69	70.55 ± 0.31	5,168,401	Tesla V100 (32GB)
LRGA + GCN Aug + Node2Vec	74.24 ± 14.18	71.54 ± 0.61	6,693,521	Tesla V100 (32GB)
LRGA + GCN Aug + Node2Vec	73.13 ± 13.79	71.73 ± 0.65	7,100,401	Tesla V100 (32GB)
LRGA + GCN Aug + Node2Vec	77.94 ± 9.20	71.81 ± 0.73	7,517,521	Tesla V100 (32GB)
LRGA + GCN Aug + Node2Vec	74.17 ± 13.97	72.03 ± 0.59	8,382,481	Tesla V100 (32GB)
LRGA + GCN Aug + Node2Vec	75.88 ± 10.28	71.65 ± 0.53	9,288,401	Tesla V100 (32GB)
LRGA + GCN Aug + Node2Vec	73.85 ± 8.71	72.25 ± 0.47	10,235,281	Tesla V100 (32GB)
LRGA + GCN Aug + Node2Vec	66.07 ± 20.60	70.97 ± 2.76	12,251,921	Tesla V100 (32GB)

Reference:

1. AGE: baseline implementation of AGE network

2. ddi: how to load ddi dataset for link prediction tasks

3. DeeperGCN: example implementation of DeeperGCN

4. LRGA: GLOBAL ATTENTION IMPROVES GRAPH NETWORKS GENERALIZATION