DTBAPred: Improved Prediction of Drug-Target Binding Affinity Using Machine Learning Approach

Abstract

Drugs are small molecules that usually bind with proteins, also called targets, to control the cellular process of target in combatting disease associated with target(s). Effectiveness of a drug hugely depends upon the strength of its binding affinity with its partner proteins. As drug discovery is a lengthy and expensive process, in silico drug discovery and drug repurposing is an alternative complementary avenue for the researchers. Nowadays drug-target binding affinity (DTBA) prediction is a part and parcel of in any in silico drug discovery and drug repurposing process. There exist many precedents in the literature which considered machine learning (ML) based approach to predict DTBA. In this thesis, we proposed a method DTBAPred which combines novel combination of features to represent drugs, targets, and feed into ML model to predict the DTBA on benchmark datasets DAVIS and KIBA. The proposed CatBoost based model outperformed some of the state-of-the-art traditional machine learning based methods in DAVIS benchmark dataset with 0.276 MSE, 0.579 R-Square, and 0.866 CI. While on KIBA benchmark dataset, the model performed at 0.219 MSE, 0.579 R-Square, and 0.835 CI. We considered two different fingerprints i.e., RDKit 2D descriptors and Morgan fingerprints to represent drugs; and Morgan fingerprints-based model showed better performance than RDKit-2D descriptors-based model, emphasizing that different fingerprints may also impact the DTBA prediction results. Feature importance analysis was conducted using SHAP, to determine the most impactful features and gain insights about the individual effect of each feature on DTBA. In summary, the obtained results indicate the superiority of the DTBAPred over several existing traditional ML models for the same purpose and emphasized the incorporation of different fingerprints in the model. We believe, our proposed method will support to improve the DTBA prediction and escalate the drug discovery process.

Date of Award	2023
Original language	American English
Awarding Institution	HBKU College of Science and Engineering