Repository for comparing different graph representations with graph convolutional neural networks. Copyright (C) 2020, Jeroen Baars, Pim Wassenaar (Contact at the RIVM in the Netherlands). This program is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version. This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. You should have received a copy of the GNU General Public License along with this program. If not, see https://www.gnu.org/licenses/. Also add information on how to contact you by electronic and paper mail. If the program does terminal interaction, make it output a short notice like this when it starts in an interactive mode: RIVM_GNN Copyright (C) 2020 Jeroen Baars.

For testing used the following repo: https://github.com/graphdeeplearning/benchmarking-gnns/tree/arXivV1

Preprocess

Benchmark_Data - create datasets for all representations
MACCS_SOMS - RF and SVM classifiers for data with and without SOMS
NN - neural networks for training
pZZS - create representations from PZZS data

Best

Contains in depth notebooks for analysis

Benchmark

Contains biogrid job submitters and GCNN code

pZZS

BestPBT -> for prediction on new dataset

#SIDENOTE (new atoms in pZZS data, problems feature vector size..)

Environments

Conda activate GNN (for preprocess)
Conda activate benchmark_gnn (for running GCNN)

Step 1

Preprocess data in benchmark data notebook or with scripts
Adjust for own dataset
Works with SMILES + TOX label spreadsheets
Parse into functions
Copy data from data folder to tox folder in benchmark

Step 2

Adjust the hardcoded parts in the benchmark dataset for new data

Step 3

Biogrid parser for running jobs
Create hyperparameter settings needed

Step 4

Parse results with notebook parses and check for best results

Step 5

Copy best to best folder and analyze results in notebook

Local

Clone repo
Conda activate GNN
Start jupyter notebook
Run Benchmarkdataset.ipynb
	For new data → adjust to own dataset
	Needed: SMILES and labels in csv/excel
Copy from preprocess/data the xxx_rep1 to xxx_rep4 data
Paste data in benchmark/data/TOX
(Adjust hardcoded parts on next page for new data)
conda deactivate GNN
conda activate benchmark_gnn
Run every representation once for 1 epoch (to create the train/test/val split)
	All splits are equal because of hardwired seed

Example for local run with python:
python main_TUs_graph_classification.py --dataset CMR_Rep4 --config ./configs/test/TOXGIN0.json
	Configs can be created with the JobSubmitterGin
Run JobSubmitterGIN once with standard settings to create config
Copy paste part from python until end and run for Rep1 to Rep4

Close notebook
Git add --all
Git commit -a -m “added representations/splits”
Git push

Biogrid

Git pull
Conda activate benchmark_gnn
Jupyter notebook
Run JobSubmitterXXX with desired grid search for hyperparameters
Check with “bjobs” in command prompt if jobs are done (4~12 hours) 
Copy “CheckResults.ipynb” to benchmark/out/tox/results
Run CheckResults and find best accuracy
Copy best checkpoints and config of accuracy to best and run particular in depth notebook
	Copy data folder from benchmark to best or pZZS folder for analysis

HARDCODED IN BENCHMARK (information for adding new dataset)

Add rep2 to list (hydro/arom rep has extra bond labels)

1 (TOX.py rule 171)

if self.name in ["PBT_Rep2", "PBT_Repn2", "CMR_Rep2"]:
	edges_dimension = 8
Else:
	edges_dimension = 4

Add list with representations and if statement with num_graphs for dataset

2 (TOX.py rule 239)

pbt = ['PBT_Rep1', 'PBT_Rep2', 'PBT_Rep3', 'PBT_Rep4']
pbtn = ['PBT_Repn1', 'PBT_Repn2', 'PBT_Repn3', 'PBT_Repn4']
cmr = ['CMR_Rep1', 'CMR_Rep2', 'CMR_Rep3', 'CMR_Rep4']

data_dir = './data/TOX'

if name in pbt:
	dataset = TOXLoad(data_dir, self.name, num_graphs=494)
elif name in pbtn:
	dataset = TOXLoad(data_dir, self.name, num_graphs=971)
elif name in cmr:
	dataset = TOXLoad(data_dir, self.name, num_graphs=652)

Add representations to list

3 (DATA.py rule 18)

toxic_reps = ['PBT_Repn1', 'PBT_Repn2', 'PBT_Repn3', 'PBT_Repn4', 'PBT_Rep1', 'PBT_Rep2', 'PBT_Rep3', 'PBT_Rep4', 'CMR_Rep1', 'CMR_Rep2', 'CMR_Rep3', 'CMR_Rep4']

# handling for (TOX) molecule dataset
if DATASET_NAME in toxic_reps:
    return TOXDataset(DATASET_NAME)