Deep learning in drug discovery: an integrative review and future challenges

Artificial Intelligence Review https://doi.org/10.1007/s10462-022-10306-1 Deep learning in drug discovery: an integrative review and future challenges Heba Askr1 · Enas Elgeldawi2 · Heba Aboul Ella4 · Yaseen A. M. M. Elshaier5 · Mamdouh M. Gomaa2 · Aboul Ella Hassanien3 Accepted: 24 October 2022 © The Author(s) 2022 Abstract Recently, using artificial intelligence (AI) in drug discovery has received much attention since it significantly shortens the time and cost of developing new drugs. Deep learning (DL)-based approaches are increasingly being used in all stages of drug development as DL technology advances, and drug-related data grows. Therefore, this paper presents a systematic Literature review (SLR) that integrates the recent DL technologies and applications in drug discovery Including, drug–target interactions (DTIs), drug–drug similarity interactions (DDIs), drug sensitivity and responsiveness, and drug-side effect predictions. We present a review of more than 300 articles between 2000 and 2022. The benchmark data sets, the databases, and the evaluation measures are also presented. In addition, this paper provides an overview of how explainable AI (XAI) supports drug discovery problems. The drug dosing optimization and success stories are discussed as well. Finally, digital twining (DT) and open issues are suggested as future research challenges for drug discovery problems. Challenges to be addressed, future research directions are identified, and an extensive bibliography is also included. Keywords Drug discovery · Artificial intelligence · Deep learning · Drug–target interactions · Drug–drug similarity · Drug side-effects · Drug sensitivity and response · Drug dosing optimization · Explainable artificial intelligence · Digital twining 1 Introduction The examination of how various drugs interact with the body and how a medication needs to act on the body to have a therapeutic impact is known as drug discovery. Drug discovery strategy constitutes from different approaches as physiology-based and target based. This strategy is based on information about the ligand and the target. In this regard, our attention was directed in certain topics especially drug (ligand)–target interactions, drug sensitivity and response, drug–drug interaction, and drug–drug similarity. For certain diseases such as cancer or pandemic situations as COVID-19, more than one drug combination is required * Aboul Ella Hassanien Extended author information available on the last page of the article 13 Vol.:(0123456789) H. Askr et al. Fig. 1  The main building blocks of the paper to alleviate the prognosis and pathogenesis interactions. Despite all the recent advances in pharmaceuticals, medication development is still a labor-intensive and costly process. As a result, several computational algorithms are proposed to speed up the drug discovery process (Betsabeh and Mansoor 2021). As DL models progress and the drug data size is getting bigger, a slew of new DLbased approaches is cropping up at every stage of the drug development process (Kim et al. 2021). In addition, we’ve seen large pharmaceutical corporations migrate toward AI in the wake of the development of DL approaches, eschewing outmoded, ineffective procedures to increase patient profit while also increasing their own (Nag et al. 2022). Despite the DL impressive performance, it remains a critical and challenging task, and there is a chance for researchers to develop several algorithms that improve drug discovery performance. Therefore, this paper presents a SLR that integrates the recent DL technologies and applications in drug discovery. This review study is the first one that incorporates the recent DL models and applications for the different categories of drug discovery problems such as DTIs, DDIs similarity, drug sensitivity and response, and drug-side effects predictions, as well as presenting new challenging topics such as XAI and DT and how they help the advancement of the drug discovery problems. In addition, the paper supports the researchers with the most frequently used datasets in the field. The paper is developed based on six building blocks as shown in Fig. 1. More than 300 articles are presented in this paper, and they are divided across these building blocks. The papers are selected using the following criteria: • The papers which published from 2000 to 2022. • The papers which published in IEEE, ACM, Elsevier, and Springer have more priority. 13 Deep learning in drug discovery: an integrative review and future… The following analytical questions are discussed and completely being answered in the paper: – AQ1: What DL algorithms have been used to predict the different categories of drug discovery problems? – AQ2: Which deep learning methods are mostly used in drug dosing optimization? – AQ3: Are there any success stories about drug discovery and DL? – AQ4: What about the newest technologies such as XAI and DT in drug discovery? – AQ5: What are the future and open works related to drug discovery and DL? The remainder of this review paper is organized as: Sect. 2 presents a review of related studies; Sect. 3 covers the various DL techniques as an overview. Section 4 presents the organization of DL applications in drug discovery problems through explaining each drug discovery problem category and gives a literature review of the DL techniques used. Section 5 discusses the numerous benchmark data sets and databases that have been employed in the drug development process. Section 6 presents the evaluation metrics used for each drug discovery problem category. The drug dose optimization, successful stories, and XAI are introduced in Sect. 7, Sect. 8, and Sect. 9. DT and open problems are suggested as future research challenges in Sects. 10 and 11. Section 12 presents a discussion of the analytical questions. Finally, Sect. 13 concludes the paper. 2 Review of related studies Although the drug discovery is a large field and has different research categories, there is a few review studies about this field and each related study has focused only on a one research category such as reviewing the DL applications for the DTIs. This section aims to review these related studies and a summary is presented in Table 1. Kim et al. (2021) presented a survey of DL models in the prediction of drug–target interaction (DTI) and new medication development. They start by providing a thorough summary of many depictions of drugs and proteins, DL applications, and widely used exemplary data sets to test and train models. One good point for this study, they identify a few obstacles to the bright future of de novo drug creation and DL-based DTI prediction. However, the major drawback of this study was that it did not consider the latest technology in DL application for the DTIs such as XAI and DTs. Rifaioglu et al. (2019) presented the recent ML applications in Virtual Screening (VS) with the techniques, instruments, databases, and m (...truncated)