DeepRES

Requirements

• python 3.10.14 (with following packages)
  • numpy 1.26.4
  • pandas 2.2.2
  • scikit-learn 1.4.2
  • pytorch 2.2.2
  • pytorch-cuda 12.1
  • pytorch-lightning 2.4.0
  • lightning 2.4.0
  • peft 0.13.2
  • transformers 4.46.3

By using environment.yml, you can build an anaconda environment exactly the same as this research.

conda env create -f environment.yml
conda activate deepres

Pretrained parameters

The model weights are available at Zenodo. Please download files and unzip.

Program Usage

EnzymeCNN

python enzymecnn.py \
    --num-inputs 20 \
    --num-channels 600 600 600 600 600 \
    --kernel-size 9 \
    --dilation 3 \
    --dropout 0.1 \
    --hidden-dims 256 \
    --epoch 2 \
    --eval-dataset data/sample_proteins.tsv \
    --mode inference \
    --checkpoint checkpoint/best_enzymecnn_checkpoint.pt \
    --outputdir result/enzymecnn/ \

• --mode: Please specify "train" or "eval" or "inference".
• --model: Please spcify "EnzymeCNN" or "EnzymeCNN-AA" or "EnzymeCNN-3Di" (default: "EnzymeCNN").

Process EnzymeCNN result

python process_enzymecnn_result.py \
    --dataset data/sample_proteins.tsv \
    --pred-scores result/enzymecnn/pred_scores.pt \
    --output result/enzymecnn/sample_proteins.processed.tsv

EnzymeCLIP

python enzymeclip.py \
    --protein-checkpoint westlake-repl/SaProt_650M_AF2 \
    --protein-use_lora \
    --reaction-checkpoint pretrained/rxnfp \
    --reaction-use_lora \
    --batch-size 128 \
    --protein-dataset data/sample_enzymes.tsv \
    --reaction-dataset data/sample_reactions.tsv \
    --outputdir result/enzymeclip/ \
    --mode inference \
    --model EnzymeCyCLIP \
    --checkpoint checkpoint/best_enzymeclip_checkpoint.ckpt

• --mode: Please specify "train" or "eval" or "inference".
• --model: Please spcify "EnzymeCLIP" or "EnzymeCyCLIP" or "EnzymeSoftCLIP" or "EnzymeSoftCyCLIP" (default: "EnzymeCyCLIP").

Process EnzymeCLIP result

python process_enzymeclip_result.py \
    --protein-dataset data/sample_enzymes.tsv \
    --reaction-dataset data/sample_reactions.tsv \
    --pred-scores result/enzymeclip/cos_sim_matrix.pt \
    --output result/enzymeclip/enzymeclip_result.tsv

Test run

You can test EnzymeCNN and EnzymeCLIP by running following commands.

EnzymeCNN

python enzymecnn.py \
    --num-inputs 20 \
    --num-channels 600 600 600 600 600 \
    --kernel-size 9 \
    --dilation 3 \
    --dropout 0.1 \
    --hidden-dims 256 \
    --eval-dataset data/sample_proteins.tsv \
    --mode eval \
    --checkpoint checkpoint/best_enzymecnn_checkpoint.pt \
    --outputdir result/testrun/enzymecnn/ \
    --seed 123

The result file is result/testrun/enzymecnn/evaluation_result.tsv:

eval_loss	eval_accuracy	eval_f1	eval_mcc	eval_auc
0.026970707811415195	0.985	0.9852216748768473	0.9704367948586523	0.9998

EnzymeCLIP

python enzymeclip.py \
    --protein-checkpoint westlake-repl/SaProt_650M_AF2 \
    --protein-use_lora \
    --reaction-checkpoint pretrained/rxnfp \
    --reaction-use_lora \
    --protein-dataset data/sample_enzymes.tsv \
    --reaction-dataset data/sample_reactions.tsv \
    --ground-truth data/sample_ground_truth.tsv \
    --mode eval \
    --model EnzymeCyCLIP \
    --checkpoint checkpoint/best_enzymeclip_checkpoint.ckpt \
    --outputdir result/testrun/enzymeclip/ \
    --gpus 1 \
    --seed 123

The result file is result/testrun/enzymeclip/enrichment_factor.tsv:

Chi	Enrichment Factor
0.05	14.703703703703702
0.1	8.407407407407407

License

DeepRES is released under the MIT License.