Deep learning has potential in the process of discovering drug, with enhanced method for analyzing image, structure of molecule and function prediction, along with preset synthesis based on the novel enzymatic structure along tailored features and its applications. Even with expanding quantity based on effective potential approaches, the statistical systems and Machine Learning algorithms that underpin them are sometimes difficult to grasp by the human mind. To meet the required recent paradigm for the automated structure of molecules, for the purpose of 'Explainable Artificial Intelligence' with deep learning approaches. In current era, there is a need for XAI with methods of deep learning to discourse the demand for a developed machine language of the molecular science. This review outlines the important concepts in XAI, possible approaches, and obstacles. It promotes to further development of XAI techniques.
TopIntroduction
In the area of biomedical environment there is a need for XAI to reach the level of explanation for drugs. Nowadays, Explainable Artificial Intelligence is very popular. XAI plays a vital role to develop “black box” approaches (Bastani O et al. 2017). The main goal behind XAI is that Artificial Intelligence programs and technologies are not understandable (Zhaoyi Chen et al. 2020). Origin of XAI comes from a set of techniques, which adds the contextual adaptability and Interpretability to Deep Learning Models. When human not able to do but Artificial Intelligence can achieve and do it, but AI models are complex for black box approaches.
This part attempts to give a complete overview of modern XAI techniques (Guang Yang et al. 2021), Figure.1 provides a review of the approaches and their targets.
The term XAI refers to Explainable Artificial Intelligence. XAI is also referred as Transparent Artificial Intelligence. XAI means how and why the algorithm makes decisions or predictions. XAI explains different cases to reach different conclusions and Strengths/Weakness of the model. XAI solves the black box models in Artificial Intelligence. XAI shows that how machines are performing their procedures and where the problem raises (Dr. Mark Roberts et al. 2019). XAI owns research in various domains, it includes, Antenna design, Algorithmic trading, Medical diagnosis, Autonomous vehicles, Text analytics, Criminal justice. There are three principles in XAI:
- 1.
Transparency addresses how a model works internally.
- 2.
Interpretability has the capability to regulate the purpose, and outcome of Deep learning model.
- 3.
XAI has the perception in representing the nodes along with the performance of model’s priority.
In this paper, Section II discusses about the background study of XAI, Drug Discovery, Explainable Artificial Intelligence and Deep learning methods, Section III describes about approaches in XAI with Drug Discovery.
TopLiterature Review
The following table contains several authors’ views about the process of drug discovery with XAI.
Table 1.
Views on the process of drug discovery using XAI
| Author | Year | Inference |
| Abhay Pandey et al. | 2020 | Briefed about the drug discovery process in different phases. |
| Arun Das et al. | 2020 | XAI algorithms, opportunities, techniques, limitations and future directions are discussed in detail. |
| Vikas K. Prajapati et al. | 2020 | This study reviews about the application of Artificial Intelligence in the process of discovering and developing drugs. |
| Amol B. Deore et al. | 2019 | Overview about the process of advanced drug discovery and its development. |
| Connor W coley et al. | 2018 | Described about the approaches based on Neural Network, and new concept is introduced to validate the chemical reaction of context in CASP. |
| Kristy A Carpenter et al. | 2018 | Due to cost and time consuming, author integrates Virtual Screening and Machine Learning in the drug discovery. |
| Andreas Mayr et al. | 2018 | Performances of the deep learning methods with drug discovery large datasets are assessed and comparison is made with target prediction and machine learning methods. |
| Lu Zhang et al. | 2017 | History of ML and advanced development of deep learning approaches and applications in RDD are summarized. |
| Himabindu Lakkaraju et al. | 2017 | BETA is a framework, it explains the behavior of black-box classifier for trustworthiness to the novel model and explained the interpretability. |
| Hughes JP et al. | 2010 | This article reviews the key stages in preclinical process of the drug discovery from Target Identification and validation through screening. |