Enzyme EC number prediction with MIL

dataset

• training set: dataset/split100.csv
• test set 1: dataset/new.csv
• test set 2: dataset/price.csv
• test set 3: dataset/uniprot-all.csv
• test set 4: dataset/uniprot-multi.csv

prepare domain instances

DCTdomain provided protein domain-level embeddings.

Here is an example.

git clone https://github.com/mgtools/DCTdomain.git
cd DCTdomain
python src/make_db.py --fafile split100.fasta --dbfile output/split100.db --gpu 1 --cpu 16

model training

Our models were trained on a single Nvidia H100 in Python 3.12 with PyTorch 2.9.

cd code
python mil-bc_emb.py --config_path config_milbc.yaml

---

pretrained models

CLEAN shared pretrained model on Google Drive

EnzHier shared pretrained model on GitHub