This repository contains code for DTI-LM, a drug-target interaction prediction framework that leverages pretrained language model and graph attention network to find representations of drugs and proteins.DTI-LM offers following advantages over existing literature.
- State-of-the-art prediction results.
- Requires minimum input data in terms of protein amino acid and drug SMILES sequences.
- Faster runtime than comparable methods.
- Investigates the current standing of language model-based DTI prediction capability and limitations.
Clone the project
git clone https://github.com/compbiolabucf/DTI-LM.gitGo to the project directory
cd DTIInstall dependencies
conda env create -f environment.ymlRun the code
python run.py --config-name bindingDB_train_GAT.yaml "tuning.param_search.tune=False" "datamodule.splitting.balanced=True" "datamodule.splitting.splitting_strategy=random"Using own data
1. Modify utils.PREPROCESS to add the data name
2. Add a preprocessing script in utils that returns
X_drug: nx1 pd.DataFrame, index=drug names, Column 1=SMILES sequence.
X_target: mx1 pd.DataFrame, index=target names, Column 1=protein sequence.
DTI: mxn (available) pd.DataFrame, index: 0-mxn, Column 1=Drug names matching X_drug index, Column 2=Target names matching X_target index, Column 3= interaction label (0,1)
3. Add a train.yaml (bindingDB_train_GAT.yaml) in configs defining preprocess and datamoduleUsing own featurizer
1. add new featurizers in module.featurizer.drug_featurizer and prot_featurizer. Return nxq and mxp embedding (p,q embedding size)
2. Modify configs.featurizer
3. Modify drug_dim and prot_dim in configs.module.GATUsing own classifier
1. Add a new pipeline in module (Similar to GAT.py and MLP.py).
2. Add necessary file in configs.module following config.module.GAT for GAT.pyHyperparameter tuning
1. Set tuning.param_search.tune=True
2. Define search space in configs.tuning
3. Modify utils.setup_config_tune to add missing hyperparameter
4. Add best hyperparameter file in config.best_params and link it to train.yaml
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent