Drug-Target Interaction (DTI) prediction using Artificial Intelligence (AI): A systematic review

Background and aims: Drug-Target Interaction (DTI) prediction is crucial in drug discovery.Knowing that drug discovery is time-consuming and costly, predicting DTI as accurately as possibleis critical. Since experimental methods are time-consuming and there is sometimes an urgentneed for a specific treatment or drug to be discovered, there should be a state-of-the-art methodfor DTI prediction. Artificial Intelligence (AI) is any system that understands the environment andcan take action based on the data comprehended to maximize the chance of achieving its goal.Artificial intelligence and various AI algorithms and models have recently been used in DTI prediction.Despite the importance of this issue, no systematic reviews or comprehensive evaluationsof studies in this field had been conducted prior to the completion of this study. Few narrativereviews have addressed this issue, mainly focusing on machine learning algorithms in a restrictednumber of databases. In this systematic review, we aimed to systematically discuss and evaluateAI-based DTI prediction, especially network-based prediction.Method: A comprehensive systematic literature search was conducted in electronic databases,including PubMed, Scopus, Embase, and Google scholar, up to October ۲۰۲۲. Two independentreviewers evaluated the retrieved publications. All studies that used AI models or algorithms topredict DTI were included. Reviews were excluded. Studies that met our inclusion criteria werethen critically appraised by two authors independently. Data such as AI algorithms and databasesused in studies were extracted.Results: We retrieved ۲۱۲ relevant publications in electronic databases. After thoroughly examiningthe titles and abstracts, removing duplicates, and in vitro studies, ۸۳ studies remained. Fulltexts of these articles were reviewed, and ultimately ۱۶ studies were included in our review. Severalalgorithms were used in the studies to detect drug-target interactions. Five of them used thekernel logistic matrix, six used ensemble learning, four used deep learning, and one used machinelearning.Conclusion: DTI prediction can be a time-consuming process. Using AI technology during thedrug discovery process can save significant time and money. In this study, we aimed to providea general taxonomy of various AI models used in DTI prediction in previous studies. Since thesestudies have only recently been proposed and developed, we see limitations and shortcomings intheir models. Despite the accurate maintenance of big data by AI algorithms shown in studies, theheterogeneity of data and shortcomings of AI algorithms can cause difficulties in using AI to predictdrug-target interactions. With further research in this field, we believe artificial intelligencecan potentially accelerate drug-targeting applications.

کلیدواژه ها:

artificial intelligence ، drug discovery ، Drug-target interaction ، AI-based DTI prediction