Thomas, Sagaya Louis GracySagaya Louis GracyThomasCHE LIN2023-07-172023-07-172022-02-0118123031https://scholars.lib.ntu.edu.tw/handle/123456789/633761Identifying drug-target interaction (DTI) is a crucial step in the drug discovery pro-cess. Several computational-based methods have been proposed to scale down the time and financial constraints imposed by the traditional methods. Although these approaches have employed various strategies to predict the interaction, they have limited the problem to a classification task that predicts the existence of interaction or a regression task that predicts the affinity of the DTI. In this study, we try to extend the regression task to a ranking task by utilizing transfer learning and listwise loss. Specifically, we utilized a model pre-trained for the regression task of DTI and used their weights for the following ranking of the top-k DTI via transfer learning. We borrowed the idea of Learn-ing-To-Rank in information retrieval and proposed a listwise ranking-based algorithm employed with the deep neural network model, which helps rank the top-k DTI. Our re-sults showed that we could improve the model’s Concordance Index from 77% to 82%, Spearman’s Correlation from 72% to 82%, and the top-10 overlap from 56% to 66% when compared to the baseline model.Deep Learning | Drug-Target Interaction | Learning To Rank | Protein Sequencejournal article10.6329/CIEE.202202_29(1).00032-s2.0-85142746059https://api.elsevier.com/content/abstract/scopus_id/85142746059