Model

1. MacroSimGNN Modification from SimGNN

To develop MacroSimGNN, we refer to and modify the source code from SimGNN (specifically, the version of be3ee61 on Jan 12, 2023 under GNU General Public License v3.0) to process macromolecule coarse-grained graph representations. The detailed modifications are described below.

1.1 Change from one-hot encoding to Morgan Fingerprint encoding

monomers_dict_full.json contains the fingerprint for each monomer symbol.

Load the Morgan fingerprint by setting path of monomers_dict_full.json line 164 in ./src/simgnn.py

        with open('path/monomers_dict_full.json','r') as json_file:

Change path to be the absolute path of monomers_dict_full.json

1.2 Add validation steps to save the model that has the minimum loss on the validation dataset

in ./src/simgnn.py

            self.model.eval()
            with torch.no_grad():
                self.test_scores = []
                self.test_ground_truth = []
                #targets = []
                #predictions = []
                for test_graph_pair in tqdm(self.testing_graphs):
                    test_data = process_pair(test_graph_pair)
                    self.test_ground_truth.append(calculate_normalized_ged(test_data))
                    test_data = self.transfer_to_torch(test_data)
                    test_target = test_data["target"]
                    test_prediction = self.model(test_data)
                    self.test_scores.append(calculate_loss(test_prediction, test_target))

            if  test_loss_epochs[-1] == min(test_loss_epochs):
                    self.save_model_state(model_state = "model_min_loss")

1.3 Add Restart command to help the model load the pre-trained model and restart.

in ./src/main.py

    if args.load_path and args.restart:
        trainer.load()
        trainer.fit()

2. Model Training and Testing

2.1 Training Command

TRAIN_DATA=~/MacroSimGNN/Dataset/train_validation_data_set/train/
VAL_DATA=~/MacroSimGNN/Dataset/train_validation_data_set/test/
SAVE_PATH=~/MacroSimGNN/Results/output_full_size/modelhistogram_new2

python ~/MacroSimGNN/Model/src/main.py \
    --epochs 1000 \
    --batch-size 256 \
    --histogram \
    --save-path $SAVE_PATH \
    --training-graphs $TRAIN_DATA \
    --testing-graphs $VAL_DATA

restart version

TRAIN_DATA=~/MacroSimGNN/Dataset/train_validation_data_set/train/
VAL_DATA=~/MacroSimGNN/Dataset/train_validation_data_set/test/
MODEL_PATH=~/MacroSimGNN/Results/output_full_size/model_min_loss
SAVE_PATH=~/MacroSimGNN/Results/output_full_size/modelhistogram_new2

python ~/MacroSimGNN/Model/src/main.py \
    --epochs 1000 \
    --batch-size 256 \
    --histogram \
    --restart \
    --load-path $MODEL_PATH \
    --save-path $SAVE_PATH \
    --training-graphs $TRAIN_DATA \
    --testing-graphs $VAL_DATA

2.2 Testing Command

TEST_DATA=~/MacroSimGNN/Dataset/test2_data_set/test/
MODEL_PATH=~/MacroSimGNN/Results/output_full_size/model_min_loss

python ~/MacroSimGNN/Model/src/main.py \
    --batch-size 256 \
    --histogram \
    --load-path $MODEL_PATH \
    --testing-graphs $TEST_DATA