All datasets are located in the data folder.
# Install from source
git clone https://github.com/abhinadduri/panspecies-dti.git
cd panspecies-dti
pip install -e .
# Or install directly from pip
install git+https://github.com/abhinadduri/panspecies-dti.git
If you want to use DDP for faster training, first follow the above installation instructions.
Then manually downgrade lightning to 2.0.8 via pip install lightning==2.0.8
Reproducing the drug-target interaction model in the MLCB 2024 abstract.
# Default config
ultrafast-train --exp-id mlcb --config configs/default_config.yaml
# Attention pooling
ultrafast-train --exp-id mlcb --config configs/agg_config.yml
The example script above will generate ProtBert and store ProtBert per-residue embeddings in a file data/BIOSNAP/full_data/train.csv.prot.h5.
The goal to start attention pooling training is to run the above script on all nested *.csv files with protein sequences in the data folder.
Links to download pre-trained models are in checkpoints/README.md.
Once downloaded, just gunzip the file to get the ready-to-use model checkpoint.
# Get target embeddings with pre-trained model
ultrafast-embed --data-file data/BIOSNAP/full_data/test.csv \
--checkpoint checkpoints/saprot_agg_contrast_biosnap_maxf1.ckpt \
--moltype target \
--output_path results/BIOSNAP_test_target_embeddings.npy
# Get drug embeddings with pre-trained model
ultrafast-embed --data-file data/BIOSNAP/full_data/test.csv \
--checkpoint checkpoints/saprot_agg_contrast_biosnap_maxf1.ckpt \
--moltype drug \
--output_path results/BIOSNAP_test_drug_embeddings.npy
ultrafast-store --data-file data/BIOSNAP/full_data/test.csv \
--embeddings results/BIOSNAP_test_drug_embeddings.npy \
--moltype drug \
--db_dir ./dbs \
--db_name biosnap_test_drug_embeddings
ultrafast-report --data-file data/BIOSNAP/full_data/test.csv \
--embeddings results/BIOSNAP_test_target_embeddings.npy \
--moltype target \
--db_dir ./dbs \
--db_name biosnap_test_drug_embeddings \
--topk 100
TODO