Investigating the Use of Ensemble Learning for Drug Target Affinity Model

Identifying the best drug candidates is important in drug research. Researchers in biology and computational science have attempted to use machine learning (ML) to predict drug-target interaction (DTI) efficiently. Most of the computational methods developed recently are to predict DTI using binary classification. However, the aim of our research is to determine whether the drug-target (DT) pair has interacted or not. Thus, we predict the DT binding affinity using the regression method, which could describe the strength of the DT pair interaction. The drug is considered not effective if the binding affinity value is not high enough. Therefore, predicting the binding affinity value is considered valuable. We propose a deep learning approach using the ensemble learning bagging method to predict the DTI. Our experiment uses CNN or CNN-RNN to encode both the drug and the protein. As for building the models, we use both CNN-CNN and CNN-RNN, which will then be tested on three publicly available datasets: BindingDB, DAVIS, and KIBA. Our research results show that by using the ensemble learning with the bagging method, we can improve the concordance index (CI) in predicting the binding affinity value by 1.36%, 1.43%, and 0.94% when using CNN-CNN for BindingDB, DAVIS, and KIBA dataset respectively.