The GO Annotation dataset can be downloaded from this link: https://current.geneontology.org/products/pages/downloads.html. Download the Homo Sapien data.

The AlphaFold v2 structure predictions can be downloaded from this link: https://alphafold.ebi.ac.uk/download#proteomes-section. Download the Homo Sapien data.

To create the environment in which to run the model, run the following:

conda env create -f environment.yml

Then activate the environment:

conda activate protein-func

Then you can do a training run with default hyperparameters by running python main.py.

Here are the runs we did:

#1 python main.py --lora_dim 16 --hidden_dim 256

#2 python main.py --model_type gat --hidden_dim 256

#3 python main.py --lora_dim 16 --hidden_dim 256 --use_conf_score False

#4 python main.py --lora_dim 16 --hidden_dim 256 --data_dir data/processed_data/hdf5_files_d_10 --feature_dim 14

#5 python main.py --lora_dim 16 --hidden_dim 256 --data_dir data/processed_data/hdf5_files_d_30 --feature_dim 34

#6 python main.py --lora_dim 16 --hidden_dim 256 --feature_dim 4

#7 python main.py --lora_dim 32 --hidden_dim 128

#8 python main.py --lora_dim 0 --hidden_dim 128 --num_layers 8

#9 python main.py --lora_dim 8 --hidden_dim 256

#10 python main.py --model_type gat --hidden_dim 512

#11 python main.py --lora_dim 32 --hidden_dim 256 --num_layers 8

#12 python main.py --lora_dim 16 --hidden_dim 384 --num_layers 8

#13 python main.py --lora_dim 32 --hidden_dim 128 --num_layers 8

#14 python main.py --model_type gat --hidden_dim 384 --num_layers 8