ArkDTA: Attention Regularization guided by non-Covalent Interactions for Explainable Drug-Target Binding Affinity Prediction

Abstract

Protein-ligand binding affinity prediction is a central task in drug design and development. Cross-modal attention mechanism has recently become a core component of many deep learning models due to its potential to improve model explainability. Non-covalent interactions, one of the most critical domain knowledge in binding affinity prediction task, should be incorporated in protein-ligand attention mechanism for more explainable deep DTI models. We propose ArkDTA, a novel deep neural architecture for explainable binding affinity prediction guided by non-covalent interactions. Experimental results show that ArkDTA achieves predictive performance comparable to current state-of-the-art models while significantly improving model explainability. Qualitative investigation into our novel attention mechanism reveals that ArkDTA can identify potential regions for non-covalent interactions between candidate drug compounds and target proteins, as well as guiding internal operations of the model in a more interpretable and domain-aware manner. (submitted to ISMB2023, under review)

Overview of ArkDTA

Attention Regularization guided by non-Covalent Interactions

Prerequisites for running ArkDTA

• Python 3.7.9
• CUDA: 11.X
• Download and extract data.tar.gz (link), 45MB) at current directory. These files are the preprocessed datasets PDBBind (ver.2020), Davis and Metz.
• Download and extract saved.tar.gz (link), 170MB) at directory ./saved. These files are the model checkpoints for each fold of the PDBbind datset.

Installing the Python (3.8.12) Conda Environment

conda env create -f arkdta.yaml
conda activate arkdta

How to use the ArkDTA source code

Training ArkDTA on PDBBind Dataset

Run the following code,

python run.py -pn {wandb_project_name} -sn arkdta -mg {multiple gpu indices}

If you want to train ArkDTA on the IC50 subset, configure the /sessions/arkdta.yaml by editing the following,

ba_measure: IC50 

Evaluating ArkDTA on PDBBind Dataset (5CV)

Run the following code,

python run.py -pn {wandb_project_name} -sn arkdta -mg {multiple gpu indices} -tm

Finetuning ArkDTA on other datasets (Davis, Metz)

Configure the /sessions/arkdta.yaml by editing the following,

dataset_subsets: davis
dataset_partition: randomsingle

Then run the following code,

python run.py -pn {wandb_project_name} -sn arkdta -mg {multiple gpu indices} -ft {davis or metz}

Evaluating ArkDTA on other datasets

Run the following code,

python run.py -pn {wandb_project_name} -sn arkdta -mg {multiple gpu indices} -tm -cn {your/saved/path_davis or _metz}

Running model inference and extracting attention maps from ArkDTA

Run the following script,

./arkdta.sh

You can change the input SMILES (ligands) or FASTA sequence (proteins) by editting the arkdta.sh file.

4x6n, 3Y5

6n77, KEJ

8bq4, QZR

Contributors

Name	Affiliation	Email
Mogan Gim	Data Mining and Information Systems Lab, Korea University, Seoul, South Korea	[email protected]
Junseok Choe	Data Mining and Information Systems Lab, Korea University, Seoul, South Korea	[email protected]
Seungheun Baek	Data Mining and Information Systems Lab, Korea University, Seoul, South Korea	[email protected]
Jueon Park	Data Mining and Information Systems Lab, Korea University, Seoul, South Korea	[email protected]
Chaeeun Lee	Data Mining and Information Systems Lab, Korea University, Seoul, South Korea	[email protected]
Minjae Ju†	LG CNS, AI Research Center, Seoul, South Korea	[email protected]
Sumin Lee†	LG AI Research, Seoul South Korea	[email protected]
Jaewoo Kang*	Data Mining and Information Systems Lab, Korea University, Seoul, South Korea	[email protected]

• †: This work was done while the author was a graduate student at Korea University Computer Science Department.
• *: Corresponding Author