A Systematic Literature Review of the Current Status and Future Prospects of Machine Learning Methods and Techniques Applied to Novel Drug Discovery

Introduction

Artificial intelligence, often known as machine intelligence, refers to a computer's ability to learn from previous data (H. Zhu, 2020). Machine Learning (ML) technologies have been and continue to be employed in a variety of sectors, including self-driving cars, improved speech recognition, medical applications, and natural language processing.

Along with the remarkable advances in high throughput sequencing technologies and genomic data explosion, artificial intelligence technologies and precisely data mining and machine learning models received an exponential interest from the scientific community in both pharmaceutical and medical research (Abdelkrim et al., 2021). The drug development industry uses computer science and artificial intelligence models to manipulate the set of data related to the construction of new medication starting from discovery and development to manufacturing, and marketing. Therefore, AI is intensively involved in the pharmaceutical sector. (Ramesh et al., 2004) by leveraging and implementing AI systems in the construction process of life-saving drugs with higher productivity, improved efficiency, and faster production (Zhang et al., 2017) as well as It could improve every aspect of manufacturing, including quality control and predictive maintenance.

Drug development is a tough and time-consuming process that typically takes between 6 and 12 years to produce a clinically safe drug that ready to be finally delivered to the market. The major risk in drug development is the uncertain result of the experiments, where despite the significant investment of money, time, and effort, there is no guarantee that the drug will be successful once it is released (Lamberti et al., 2019). Drug discovery and design is the process of searching for and identifying new medicine. Identifying new drugs is an extremely long and tedious process which is based on different approaches and strategies. The drug discovery process typically begins with the identification of a specific disease and the definition of the unmet medical needs, followed by the identification and validation of a druggable target molecule through the development of both in-vivo and in-vitro trials. Many molecules can be processed and then selected to select a small number of potential candidates. However, the main problem is that the quantity of ligands to be tested has risen dramatically; we are currently taking into account more than 10^13 ligands. Eventually, considerable efforts have been made to minimize the process's cost and duration while boosting the drug effectiveness (DiMasi et al., 2016).

As an effective solution to manage the explosion of the number of ligands, Artificial intelligence (AI) has been used to speed up drug development by combining a variety of data sources, such as publicly available datasets, databases, and various types of textual material, into a knowledge ground that is used to train machine learning models at different stages of the drug development and discovery process (Vamathevan et al., 2019).

In this paper, we explore and discuss how machine learning can be used in the different steps of drug discovery and development. It should be emphasized that machine learning methods have been extensively used in literature. As a result, we chose for this article papers exhibiting highly novel ML/DL results. In addition, abstracts, posters, and technical reports were excluded from the study. Following the findings of an initial collection of relevant articles, we first removed duplicates from multiple sources and then did a thorough examination to evaluate the remaining articles based on the inclusion criteria: the paper is original and full-text, and focuses on machine learning applications and deep learning approaches in drug discovery and development.

The remainder of the paper is organized as follows: Section 2, presents some of the most important concepts in drug discovery and development. In section 3, we introduce some theoretical background on machine learning and its application to drug discovery as well as some of the most cited databases in the field. Section 4, is a thorough summary of the most significant contributions in the field. The paper discusses the findings and presents the major challenges, and prospects future directions in section 5. Finally, section 6 closes our study with general remarks and our future research perspectives.