Available online on 15.01.2023 at http://jddtonline.info

Journal of Drug Delivery and Therapeutics

Open Access to Pharmaceutical and Medical Research

Copyright  © 2023 The  Author(s): This is an open-access article distributed under the terms of the CC BY-NC 4.0 which permits unrestricted use, distribution, and reproduction in any medium for non-commercial use provided the original author and source are credited

Open Access   Full Text Article                                                                                                                                                                            Review Article 

Application of Artificial Intelligence and Machine Learning in Drug Discovery and Development

Madhukiran Parvathanenia* , Abduselam K. Awola, Monika Kumaria,  Ke Lana, Manisha Lingama 

aHarrisburg University of Science and Technology, Harrisburg, Pennsylvania, USA 17101

Figure 1: The application of AI, ML and DL in therapeutics development

One of the fundamental goals of mankind is to regulate periodic changes for our benefit. This is certainly relevant in the fields of medicine and pharmaceuticals, which are frequently undergoing change. The creation of new ideas or interpretations in general chemical and mechanical engineering has been driven by the drug and biopharmaceutical sectors, which have been a restricted source of imaginative and original technologies or machinery. Mechanical innovation is urgently needed in the pharmaceutical sector to facilitate the production of drugs for human use. The "one size fits all" philosophy underlies current prescription practices; nonetheless, numerous crucial fields of medicine demand fresh approaches, necessitating new pharmaceutical development procedures. As a result, Artificial Intelligence (AI) is being used more frequently, which will potentially alter the ways in which clinical evaluations and trainings are conducted as well. A modern concept known as "big data" describes expansive datasets that are insurmountable via traditional data analysis; the rapid deployment of AI-enabled technology on these large clinically derived datasets continues to provide useful information that can inform the course of drug development1.

Traditional drug discovery is often an expensive, laborious, and often time-consuming pursuit in which there exists a large chasm between the identification of promising drug candidates and the employment of such novel drugs in the clinic2. As such, there has been an increasing interest in leveraging artificial intelligence (AI) and machine learning (ML)(as illustrated by Figure 1) to expedite drug discovery as well as drug development, in addition to reducing the associated financial costs. Drug discovery is loosely divided into 1) identification of disease of interest and the molecular drug target as well as its validation, 2) development of a high throughput invitro assay to screen for compound hits against the identified target and 3) optimization of potency and selectivity of hit compounds to enable in vivo testing3. Currently AI is most often used in the identification and validation of target proteins as well as in the development and optimization of hit molecules and drug candidates4. On the other hand, drug development is often divided into the preclinical stage and the clinical stage. In the preclinical stage in vivo testing in animals is conducted to assess toxicity, safety, stability (summarized as Structure-Activity Relationship (SAR) as well as, manufacturing optimization and clinical trial protocol drafting. In the clinical stage, the efficacy and safety of the novel drug is assessed in humans after the approval of the IND (investigational New Drug) application3 as shown in Figure 2 and Figure 4.

The introduction of AI and ML into the pharmaceutical industry has resulted in the entrance of large non-pharmaceutical tech companies into the health sector such as IBM, Microsoft, and Google. Novartis has started to collaborate with Microsoft and has developed a new way to customize drugs. Nvidia has also established a similar partnership with Schrodinger in early 2021; the coalescence of expertise in the design of GPU software with expertise in drug development, shall yield biomolecular predictions that are of high accuracy and high performance. Historically, drug discovery as well as development has taken an average of 12-15 years and has cost billions of dollars. But, the integration of AI and Machine Learning, is a promising avenue that can greatly reduce the time and monetary demands of the drug development process. This emerging technology will not only help to develop more effective drugs but will also help to develop drugs with high specificity toward their target. Additionally, the expertise required in integrating AI and ML into the drug discovery and development pipeline will continue to require partnerships between pharmaceutical companies and data science-based technology companies. There is a growing interest in 100% in silico drug development according to leaders in the field such as the CEO and President of IKTOS, Yann

Gaston-Mathe in addition to, projections of increased market value of companies that conduct AI-integrated high throughput screening due to an increase in the efficacy of hits1.

Figure 2: Broad stages of drug development prior to FDA approval

Proteins are critical molecules in the human body and are key in mediating the bioactivity of drugs and certain conformational changes can lead to dysfunction and then disease. The 3-D structure of a protein can provide insights about how a ligand, or a drug would engage it. There has also been a dearth in libraries of the 3-D structure of large molecules as they have been difficult to resolve through biological experimentation, but the structures of small molecules that could be optimized into drugs have been successfully predicted via ligand binding predicting programs such as Alphafold (that have revolutionized the target identification stage of drug discovery).

Through leveraging its deep learning neural network (a form of machine learning), Alphafold can predict the 3-D structure of protein/target, as well as the associated target binding affinity; the target is often the most significant causal factor for a disease of interest5. The AI computer system IBM Watson has also discovered RNA-binding proteins responsible for ALS(Amyotrophic Lateral Sclerosis), and has reported 4 prominent RNA-binding proteins worth targeting6,7.There has also arisen a great potential for the further exploration of small molecule conjugated antibody therapeutics as a result of growing protein structure databases, due to their unique target binding affinity8.

The elucidation of the structure of a target protein is essential in designing an effective drug and AI and ML has been increasingly used in the predication of protein conformational structure to identify druggable protein targets9. Recently, the frontrunner, protein structure predicting AI program and winner of 2020’s CASP(Critical Assessment of Structure Prediction) competition has been Alphafold which also inspired offshoots such as RoseTTAFold10. DeepMind’s Alphafold can leverage experimentally determined protein structures and amino acid sequences from the vast Protein Data Bank(PDB) to glean evolutionary relationships between amino acids to predict the three dimensional structure of related proteins through its deep learning networks(a type of machine learning that emulates human neural networks); the predicted structures also receive a score which reflects the program’s assessment of the accuracy of the structural output11. Nonetheless, Alphafold also has some significant blind spots such as its inability to account for protein folding pathways and doesn’t account for the different conformational structures that arise upon changes in activation or even upon post-translational modification; this limits Alphafold to a tool best used in early drug discovery as a supplemental tool to experimental determination of protein structure12.Target validation is a crucial step of drug discovery and as such AI and ML based in-silico validation requires training with high-quality and rich data sets13; this is key as ineffective, toxic and non-safe compounds need to be filtered out and must not be allowed to carry on in the development pipeline. The growing power of protein structure predicting AI programs is promising but cannot be completely depended on for target validation as of yet.

Hit identification is the next step in drug discovery and in this process large swaths of molecules are exposed to the target to identify molecules that bind with greater affinity. Deep Docking is a QSAR model dependent, deep learning tool, that allows for the identification of the best hit compounds by thousands of times greater enrichment of compound without the loss of any effective binders14. Similarly, iterative screening also helps in extraction of the most promising compound from compound libraries and facilitates high-throughput screening. There are various machine learning methods being used for hit identification and screening such as RF (Random Forest), Support Vector Machines (SVM), Light Gradient Boosting Machine (LGBM), Deep Neural Network (DNN) and Graph Convolutional Neural Network15,16.

The demand for in silico technologies or AI applications in pharmaceutical development research has increased and will continue to do so, due to a need for accurate prediction of pharmacokinetics/ADMET (Absorption, Distribution, Metabolism and Toxicity, as shown in Figure 3) of hit compounds, as pharmacokinetics as well as toxicity need to be assessed to prevent the failure of candidate drug in a later stage of drug development. The pharmacokinetics of a molecule can be predicted and modeled in relation to the target protein’s 3-D structure, through various methods like molecule docking, dynamics simulation, quantum mechanisms as well as PBPK (Quantitative-activity relationship and physiologically based pharmacokinetic) modeling17. Between 2005 and 2010, Astra Zeneca worked on small molecule drug discovery and found that most drugs failed in development due to their toxicity. Through machine learning they found that they could have predicted ADMET at an earlier stage and could have saved a lot of money and time as well. AI not only allows for the high throughput optimization of lead compounds but also makes it possible to assess target affinity that can be optimized via medicinal chemistry18. Lipinski “rule of five”, which has been used to assess drug-likeness and the likelihood of oral bioavailability of drugs in development, has at times been a barrier for advancing drugs with drug-likeness properties but violate a Lipinski rule and AI can be applied to resolve their drug candidacy through machine learning at a lower time and financial cost19,20.

Figure 3: ADMET properties

Figure 4: Application of AI and ML at different stages of drug development

The issues related to traditional computational drug development methods were those of computational limitations and of high time consumption. However, traditional computational drug development still played an essential role in drug discovery but has greatly benefited from the integration of AI as well as ML which have greatly reduced the time and financial demands as shown in Figure 4. Today, ML enabled tools are revolutionizing ligand-binding target protein identification based on 3-D structure with the prime example being DeepMind’s AlphaFold, which as described earlier predicts 3-D protein structures through amino acids sequences. Such advancement in drug development will allow for speedy design of drugs with high binding affinities for their target proteins. There are several areas where traditional drug development has been improved by AI integration and has resulted in FDA approved drugs; the reduced processing time in molecular dynamics simulation by CNN-AI enabled, de novo drug simulating designing tool, MolAlCal and multiparametric QSAR are prime examples21.

QSAR(Quantitative Structure-Activity Relationship) is a form of SAR in which statistical and computational modeling that is bolstered by AI and ML is employed to predict the biological and physical properties as well as the activity of novel drugs, in a cost and time efficient manner; this varies from traditional SAR which entails exhaustive and costly biological assays15,22 . Any form of predictive computational modeling is only as good as its parameters and as such there are multiple types of QSAR methods that are classified based on the parameters they consider and can be utilized in varying circumstances depending on the parameter of interest23. QSAR is integral in drug discovery as it allows for the generation of a small pool of molecules of interest that may possess desirable properties that can be further investigated and ascertained in vivo, and this is especially useful when screening large compound libraries as it ameliorates the significant time and resource burden that plagues early traditional drug discovery24. Classical QSAR(which refers to the Hansch and Free-Wilson analysis methods) operates under the assumption that the chemical structure of compounds determines their bioactivity, and leverages an input of a library of molecules that includes their physical chemistry properties, known bioactivity, chemical structures as well as the known activity of the ligands of interest to conduct statistical relationship assessments that can be extrapolated to predict the bioactivity and physical chemistry properties of a novel compound without the need for organic syntheses15,25,26. QSAR methods have matured since their introduction in the 1960s by Corwin Hansch and have grown to include the 3D structure of molecules/ligands which allows for more accurate prediction of ligand-receptor interactions, as well as the determination of relationship between a ligand’s structure and bioactivity; this method is known as 3D-QSAR and is the most widely used QSAR method today27,28,29. As idyllic as QSAR may sound, it isn’t without its weaknesses and in no way could supplant traditional SAR30.

ML has been of great interest to drug developers and data scientists alike and there are various ML methods that are in use today such as LR, RF, and SVM, as previously mentioned15,16,31. However, all these methods suffer from a decrease in accuracy when it comes to molecular prediction of complex compounds and have resulted in a more complex form of ANN (Artificial Neural Network) known as Deep Learning (DL), which is multilayered by design. After the input layer (which takes in the data to be processed), the following layers of a DL network are trained by each preceding layers’ results; as a result this network can process large amounts of data that it can use to conduct complex computational modeling and extrapolate to predict the structure of more complex molecules through the non-linearity function. Other examples of ANN are CNN(Convolutional Neural Networks) and SNN (Self-Normalizing Neural Networks)31as shown in Figure 5. Nowadays, the combination of Perturbation Theory with ML which results in PTML, is being used for modeling toxicity of drugs in varying cellular environments as well as in varying experimental animal models. PTML modelling, can help drug developers to understand why a drug may have failed in in-vivo studies, but also could help to remodel the chemical structure of the drug to reduce toxicity during the preclinical trials; PTML modeling could also help to increase the small number of drugs that are screened and progress to Phase 1 trials. Most of the hits in early drug discovery are excluded during the hit to lead optimization, as well as pharmacokinetics and toxicity assessments, and PTML modeling could help to identity the best drug candidates sooner and more efficiently. Initially withdrawn hits can now be structurally modified via ML modeling, in which the changes in toxicity can be predicted as a result of structural modification, also providing insights such as organ specificity, carcinogenicity, genotoxicity, and LD₅₀ quantification. Computational modeling also increases the chance that novel drugs that may be toxic to animal models do not proceed to animal studies making the drug development process one that is more humane. Furthermore, computational modeling such as PTML modeling allows for the development of drugs with higher efficacy and lower toxicity in a time efficient and resource efficient manner31.

Figure 5: Application of AI and ML in the generation of FDA approved drugs

Recent years have witnessed cloud data centers (DCs) gaining significant prominence when it comes to performing data management and shared computing. The high utilization of shared resources has already exhibited several advantages, but it needs to be acknowledged that data often comes in an unstructured format, and this is where the application of machine learning algorithms steps in. Performing reliable resource provisioning which can operate across multiple platforms is not an easy accomplishment. Despite the presence of conflicting conjectures regarding the resource efficiency of machine learning, it is hard to ignore the impressive attributes of machine learning algorithms when it comes to performing prediction, classification, and forecasting32. As has been identified by Hussain et al., traditional resource management techniques have become outdated and now possess a wide range of limitations33.With the continual growth in the implementation of IoT devices, a plethora of data is being generated on a regular basis; the application of proper machine learning techniques can harness the growing supply of information associated with IoT data to more effectively allocate resources and generate predictions34.

According to Murali et al., several attempts have been made over the years to integrate automation in the task of resource allocation and it has been deduced that machine learning is one of the optimum solutions35. It has also been acknowledged that DNN (Distributed Neural Networks) are the future of automation and that it is the area which possesses the potential to perform computation on large databases effortlessly36. Additionally, it has also been observed that the resource utilization system of vehicular networks has grown dynamically with the advent of AI based resource allocation systems; the application of machine learning techniques has also resulted in improvements in vehicular resource allocation strategy over time37. The operations involving resource allocation have supplementally become much less volatile due to the presence of historical data and the rapid nature of machine learning algorithms, which simplifies the assortment of resources38.

Big data within healthcare has introduced the usual aspects of volume, velocity, and veracity in maintenance and management of patient data39. Machine Learning tools generally harvest the accumulated data which are generally structured, thus neural networks or naive Bayesian approaches have been implemented within the field of data science to accumulate the most probable outcomes of queries that are posed to the dataset40(as shown in Figure 4 and Figure 5).

Machine learning techniques have increasingly implicated in drug formulation in the pharmaceutical industry and can help synthesize medical drugs according to Elbadawi et al.41. M3DISEEN is an example of such an application and has been a web-based pharmaceutical application that initiates the FDM 3D printing process for developed drugs through formulations stocks. The modern advent of ML data harvesting techniques also allows the use of machine learning techniques that can improve drug discovery probabilities for CNS-related diseases; this machine learning technique has also helped in the drug actuation process where designing and activity measurement can be performed as well42. The application of machine learning can also improve the acceptability of 3D-printed drugs further enhancing the quality control standards of traditionally manufactured drugs as well43.

A new paradigm of genetic data collection has enabled early cancer risk detection within non- diseased individuals, and the detection of antigens from the cancerous individuals has helped researchers to develop combinatorial antibodies to fight against the disease onset as well44. Gut microbiomes provide a levy of genetic databases which can later be tallied according to the geographical variables with the help of machine learning techniques. Thus in-silico preparation of microbiome-specific drugs can help patients to recover from their ailments45. Specific gut microbiome combinations have been tallied using artificial intelligence and drugs are suggested accordingly. Finally, it can be said that the application opportunities for these machine learning techniques in the formulation of drugs have been immensely diverse, but these techniques along with fuzzy logic and genetic algorithms will continue to streamline both drug design and drug formulation and manufacturing46.

Drug discovery and development continues to be a long and arduous road that begins with billions of compounds that are narrowed down based on their target engagement, binding affinities and ADMET/pharmacokinetic properties. The integration of AI and ML into the drug discovery allows for further investigation of withdrawn drug candidates may have either violated the Lipinski rule of 5(as shown in Figure 6) or failed toxicity screening and ML modeling makes it possible to modify compounds’ structure to reduce their toxicity. Modified drug candidates can then be easily assessed for their efficacy and ADMET properties (as shown in Figure 3) through DL. Not only does AI enabled computational modeling allow for more efficient drug discovery and development but also could make preclinical studies in animal studies more humane by helping to withdraw toxic drug candidates.

Figure 6: Lipinski rule of 5

QSAR modeling continues to evolve and develop but currently falls short of being able to supplant SAR entirely due to its declining accuracy with increasing compound complexity.

Nonetheless, QSAR methods have developed to move past receptor independent models into receptor-dependent QSAR(RD-QSAR) which uses the multiple 3D structures of the ligand- receptor conformations to allow for the prediction of the conformations of other ligand-receptor interactions (RD-4D-QSAR) allowing for more detailed predictions of how a drug of interest may interact with the target of interest; both receptor dependent and receptor independent QSAR has recently been reinvigorated and has been successfully applied in drug discovery due to the continuing growth in computational and graphical processing power47,48. There has also been an increased interest in the use of deep learning QSAR, which would suit the multi parametric nature of drug optimization, but it suffers from low robustness due to the massive computational power and large amounts of data that are necessary for accurate prediction generation; however the rise in the incorporation of uncertainty estimation into neural networks may provide correction factors to incorrect property predictions that are a symptom low data based prediction models49. Alphafold has also gained a new rival in the protein structure prediction space in the form of Meta AI’s ESMFold which has characterized the structure of about 600 million proteins derived from microorganisms that have not been previously characterized50. Despite being faster at predicting protein structures than Alphafold in certain cases, ESMFold is not as accurate as Alphafold but distinguishes itself in its more structurally unique and rich metagenomic protein prediction database as well as having a prediction model that is better suited for more easily determining the structural impact of mutations.

The utilization of AI and ML in drug discovery and development (especially in the development of oncological and psychiatric drug) in FDA submissions has been on the rise since the 2010s and as such the FDA has had to explore how this nascent form of therapeutics development, administration and selection is to be regulated51.The FDA released a discussion paper titled, “Proposed Regulatory Framework for Modifications to Artificial Intelligence/Machine Learning(AI/ML)-Based Software as a Medical Device(SaMD)-Discussion Paper and Request for Feedback”, in 2019 which primarily aimed to ensure that users were made aware of self-learning and machine learning based modifications and developments and that risk- assessment of these modifications was administered throughout the FDA approval process and beyond52. The 2019 discussion paper addressed the use of AI and ML as a medical device which limits the regulation discussion to its use in the clinic (when it may be used in treatment and in aiding diagnostics) and there does not seem to be any existing regulatory advisement on the use of AI and ML in drug discovery and development53. Such regulatory advisements may not have been introduced by the FDA as findings that direct the course of drug discovery and drug development as AI and ML are always validated by biochemical assays and invitro studies prior to progressing in the pipeline. However, the FDA has been promoting the use Real-World Data (RWD) since the mid 2010s, in an effort to encourage the collection and analysis of extensive clinical medical information from clinical study patients to better inform drug development and has resulted in the employment of AI for tracking the adverse effects of drugs in the clinic (also known as pharmacovigilance) as well as in investigating a drug’s potential for repurposing54,55.

Conflict of Interest-

All authors of this manuscript declare that this publication is void of any conflicts of interest.

Acknowledgement-

The authors are grateful to Madhukiran Parvathaneni, PhD for his guidance in the composition of this review manuscript.

References

1. Savage N. Tapping into the drug discovery potential of AI. Biopharm Deal. 2021 May 27; d43747-021-00045-7. https://doi.org/10.1038/d43747-021-00045-7

2. Chan HCS, Shan H, Dahoun T, Vogel H, Yuan S. Advancing Drug Discovery via Artificial Intelligence. Trends in Pharmacological Sciences. 2019 Aug; 40(8):592-604. https://doi.org/10.1016/j.tips.2019.06.004

3. Sinha S, Vohora D. Drug Discovery and Development. In: Pharmaceutical Medicine and Translational Clinical Research [Internet]. Elsevier; 2018. p. 19-32. Available from: https://linkinghub.elsevier.com/retrieve/pii/B978012802103300002X https://doi.org/10.1016/B978-0-12-802103-3.00002-X

4. Jung YL, Yoo HS, Hwang J. Artificial intelligence-based decision support model for new drug development planning. Expert Systems with Applications. 2022 Jul; 198:116825. https://doi.org/10.1016/j.eswa.2022.116825

5. Schauperl M, Denny RA. AI-Based Protein Structure Prediction in Drug Discovery: Impacts and Challenges. J Chem Inf Model. 2022 Jul 11; 62(13):3142-56. https://doi.org/10.1021/acs.jcim.2c00026

6. Mak KK, Pichika MR. Artificial intelligence in drug development: present status and future prospects. Drug Discov Today. 2019 Mar; 24(3):773-80. https://doi.org/10.1016/j.drudis.2018.11.014

7. Schenone M, Dančík V, Wagner BK, Clemons PA. Target identification and mechanism of action in chemical biology and drug discovery. Nat Chem Biol. 2013 Apr; 9(4):232-40. https://doi.org/10.1038/nchembio.1199

8. Srivastava R. Theoretical Studies on the Molecular Properties, Toxicity, and Biological Efficacy of 21 New Chemical Entities. ACS Omega. 2021 Sep 28;6(38):24891-901. https://doi.org/10.1021/acsomega.1c03736

9. Mohs RC, Greig NH. Drug discovery and development: Role of basic biological research. Alzheimer's & Dementia: Translational Research & Clinical Interventions. 2017 Nov; 3(4):651-7. https://doi.org/10.1016/j.trci.2017.10.005

10. Mullard A. What does AlphaFold mean for drug discovery? Nat Rev Drug Discov. 2021 Oct; 20(10):725-7. https://doi.org/10.1038/d41573-021-00161-0

11. Callaway E. What's next for AlphaFold and the AI protein-folding revolution. Nature. 2022 Apr 14; 604(7905):234-8. https://doi.org/10.1038/d41586-022-00997-5

12. Nussinov R, Zhang M, Liu Y, Jang H. AlphaFold, Artificial Intelligence (AI), and Allostery. J Phys Chem B. 2022 Sep 1; 126(34):6372-83. https://doi.org/10.1021/acs.jpcb.2c04346

13. Aittokallio T. What are the current challenges for machine learning in drug discovery and repurposing? Expert Opinion on Drug Discovery. 2022 May 4; 17(5):423-5. https://doi.org/10.1080/17460441.2022.2050694

14. Gentile F, Yaacoub JC, Gleave J, Fernandez M, Ton AT, Ban F, et al. Artificial intelligence-enabled virtual screening of ultra-large chemical libraries with deep docking. Nat Protoc. 2022 Mar; 17(3):672-97. https://doi.org/10.1038/s41596-021-00659-2

15. Carracedo-Reboredo P, Liñares-Blanco J, Rodríguez-Fernández N, Cedrón F, Novoa FJ, Carballal A, et al. A review on machine learning approaches and trends in drug discovery. Computational and Structural Biotechnology Journal. 2021; 19:4538-58. https://doi.org/10.1016/j.csbj.2021.08.011

16. Dreiman GHS, Bictash M, Fish PV, Griffin L, Svensson F. Changing the HTS Paradigm: AI-Driven Iterative Screening for Hit Finding. SLAS Discovery. 2021 Feb; 26(2):257-62. https://doi.org/10.1177/2472555220949495

17. Jia L, Gao H. Machine Learning for In Silico ADMET Prediction. In: Heifetz A, editor. Artificial Intelligence in Drug Design [Internet]. New York, NY: Springer US; 2022 [cited 2022 Dec 5]. p. 447-60. (Methods in Molecular Biology; vol. 2390). Available from: https://link.springer.com/10.1007/978-1-0716-1787-8_20 https://doi.org/10.1007/978-1-0716-1787-8_20

18. Wu F, Zhou Y, Li L, Shen X, Chen G, Wang X, et al. Computational Approaches in Preclinical Studies on Drug Discovery and Development. Front Chem. 2020 Sep 11; 8:726. https://doi.org/10.3389/fchem.2020.00726

19. Vijayan RSK, Kihlberg J, Cross JB, Poongavanam V. Enhancing preclinical drug discovery with artificial intelligence. Drug Discovery Today. 2022 Apr; 27(4):967-84. https://doi.org/10.1016/j.drudis.2021.11.023

20. Walters WP. Going further than Lipinski's rule in drug design. Expert Opinion on Drug Discovery. 2012 Feb; 7(2):99-107. https://doi.org/10.1517/17460441.2012.648612

21. Gupta R, Srivastava D, Sahu M, Tiwari S, Ambasta RK, Kumar P. Artificial intelligence to deep learning: machine intelligence approach for drug discovery. Mol Divers. 2021 Aug; 25(3):1315-60. https://doi.org/10.1007/s11030-021-10217-3

22. Muratov EN, Bajorath J, Sheridan RP, Tetko IV, Filimonov D, Poroikov V, et al. QSAR without borders. Chem Soc Rev. 2020; 49(11):3525-64. https://doi.org/10.1039/D0CS00098A

23. Verma J, Khedkar V, Coutinho E. 3D-QSAR in Drug Design - A Review. CTMC. 2010 Jan 1; 10(1):95-115. https://doi.org/10.2174/156802610790232260

24. Kwon S, Bae H, Jo J, Yoon S. Comprehensive ensemble in QSAR prediction for drug discovery. BMC Bioinformatics. 2019 Dec; 20(1):521. https://doi.org/10.1186/s12859-019-3135-4

25. Van Valkenburg W, editor. Biological Correlations-The Hansch Approach [Internet]. WASHINGTON, D. C.: AMERICAN CHEMICAL SOCIETY; 1974 [cited 2022 Dec 5]. (Advances in Chemistry; vol. 114). Available from: https://pubs.acs.org/doi/book/10.1021/ba-1972-0114 https://doi.org/10.1021/ba-1972-0114

26. Cronin MTD, Schultz TW. Pitfalls in QSAR. Journal of Molecular Structure: THEOCHEM. 2003 Mar; 622(1-2):39-51. https://doi.org/10.1016/S0166-1280(02)00616-4

27. Clark R. Prospective Ligand- and Target-Based 3D QSAR: State of the Art 2008. CTMC. 2009 Jun 1; 9(9):791-810. https://doi.org/10.2174/156802609789207118

28. Cruz VL, Martinez S, Ramos J, Martinez-Salazar J. 3D-QSAR as a Tool for Understanding and Improving Single-Site Polymerization Catalysts. A Review. Organometallics. 2014 Jun 23; 33(12):2944-59. https://doi.org/10.1021/om400721v

29. Jagiello K, Grzonkowska M, Swirog M, Ahmed L, Rasulev B, Avramopoulos A, et al. Advantages and limitations of classic and 3D QSAR approaches in nano-QSAR studies based on biological activity of fullerene derivatives. J Nanopart Res. 2016 Sep; 18(9):256. https://doi.org/10.1007/s11051-016-3564-1

30. Paul D, Sanap G, Shenoy S, Kalyane D, Kalia K, Tekade RK. Artificial intelligence in drug discovery and development. Drug Discovery Today. 2021 Jan; 26(1):80-93. https://doi.org/10.1016/j.drudis.2020.10.010

31. Pérez Santín E, Rodríguez Solana R, González García M, García Suárez MDM, Blanco Díaz GD, Cima Cabal MD, et al. Toxicity prediction based on artificial intelligence: A multidisciplinary overview. WIREs Comput Mol Sci [Internet]. 2021; 11(5). Available from: https://onlinelibrary.wiley.com/doi/10.1002/wcms.1516 https://doi.org/10.1002/wcms.1516

32. Sharma SK, Wang X. Toward Massive Machine Type Communications in Ultra-Dense Cellular IoT Networks: Current Issues and Machine Learning-Assisted Solutions. IEEE Commun Surv Tutorials. 2020; 22(1):426-71. https://doi.org/10.1109/COMST.2019.2916177

33. Hussain F, Hassan SA, Hussain R, Hossain E. Machine Learning for Resource Management in Cellular and IoT Networks: Potentials, Current Solutions, and Open Challenges. 2019; Available from: https://arxiv.org/abs/1907.08965

34. Yang H, Alphones A, Xiong Z, Niyato D, Zhao J, Wu K. Artificial-Intelligence-Enabled Intelligent 6G Networks. IEEE Network. 2020 Nov; 34(6):272-80. https://doi.org/10.1109/MNET.011.2000195

35. Murali A, Das NN, Sukumaran SS, Chandrasekaran K, Joseph C, Martin JP. Machine Learning Approaches for Resource Allocation in the Cloud: Critical Reflections. In: 2018 International Conference on Advances in Computing, Communications and Informatics (ICACCI) [Internet]. Bangalore: IEEE; 2018; 2073-9. Available from: https://ieeexplore.ieee.org/document/8554703/

https://doi.org/10.1109/ICACCI.2018.8554703

36. Toczé K, Nadjm-Tehrani S. A Taxonomy for Management and Optimization of Multiple Resources in Edge Computing. Wireless Communications and Mobile Computing. 2018 Jun 4; 2018:1-23. https://doi.org/10.1155/2018/7476201

37. Nurcahyani I, Lee JW. Role of Machine Learning in Resource Allocation Strategy over Vehicular Networks: A Survey. Sensors. 2021 Sep 30; 21(19):6542. https://doi.org/10.3390/s21196542

38. Wang J, Jiang C, Zhang H, Ren Y, Chen KC, Hanzo L. Thirty Years of Machine Learning: The Road to Pareto-Optimal Wireless Networks. IEEE Commun Surv Tutorials. 2020; 22(3):1472-514. https://doi.org/10.1109/COMST.2020.2965856

39. Musella S, Verna G, Fasano A, Di Micco S. New Perspectives on Machine Learning in Drug Discovery. CMC. 2021 Oct 15; 28(32):6704-28. https://doi.org/10.2174/0929867327666201111144048

40. Damiati SA. Digital Pharmaceutical Sciences. AAPS PharmSciTech. 2020 Aug; 21(6):206. https://doi.org/10.1208/s12249-020-01747-4

41. Elbadawi M, Muñiz Castro B, Gavins FKH, Ong JJ, Gaisford S, Pérez G, et al. M3DISEEN: A novel machine learning approach for predicting the 3D printability of medicines. International Journal of Pharmaceutics. 2020 Nov; 590:119837. https://doi.org/10.1016/j.ijpharm.2020.119837

42. Vatansever S, Schlessinger A, Wacker D, Kaniskan HÜ, Jin J, Zhou M, et al. Artificial intelligence and machine learning‐aided drug discovery in central nervous system diseases: State‐of‐the‐arts and future directions. Med Res Rev. 2021 May; 41(3):1427- 73. https://doi.org/10.1002/med.21764

43. Elbadawi M, McCoubrey LE, Gavins FKH, Ong JJ, Goyanes A, Gaisford S, et al. Disrupting 3D printing of medicines with machine learning. Trends in Pharmacological Sciences. 2021 Sep; 42(9):745-57. https://doi.org/10.1016/j.tips.2021.06.002

44. Iqbal MJ, Javed Z, Sadia H, Qureshi IA, Irshad A, Ahmed R, et al. Clinical applications of artificial intelligence and machine learning in cancer diagnosis: looking into the future. Cancer Cell Int. 2021 Dec; 21(1):270. https://doi.org/10.1186/s12935-021-01981-1

45. McCoubrey LE, Elbadawi M, Orlu M, Gaisford S, Basit AW. Harnessing machine learning for development of microbiome therapeutics. Gut Microbes. 2021 Jan 1; 13(1):1872323. https://doi.org/10.1080/19490976.2021.1872323

46. Sethuraman N. Artificial Intelligence: A New Paradigm for Pharmaceutical Applications in Formulations Development. IJPER. 2020 Dec 22; 54(4):843-6. https://doi.org/10.5530/ijper.54.4.176

47. Andrade CH, Pasqualoto KFM, Ferreira EI, Hopfinger AJ. 4D-QSAR: Perspectives in Drug Design. Molecules. 2010 May 4; 15(5):3281-94. https://doi.org/10.3390/molecules15053281

48. Bak A. Two Decades of 4D-QSAR: A Dying Art or Staging a Comeback? IJMS. 2021 May 14; 22(10):5212. https://doi.org/10.3390/ijms22105212

49. Jiménez-Luna J, Grisoni F, Weskamp N, Schneider G. Artificial intelligence in drug discovery: recent advances and future perspectives. Expert Opinion on Drug Discovery. 2021 Sep 2; 16(9):949-59. https://doi.org/10.1080/17460441.2021.1909567

50. Callaway E. AlphaFold's new rival? Meta AI predicts shape of 600 million proteins. Nature. 2022 Nov 10; 611(7935):211-2. https://doi.org/10.1038/d41586-022-03539-1

51. Liu Q, Huang R, Hsieh J, Zhu H, Tiwari M, Liu G, et al. Landscape Analysis of the Application of Artificial Intelligence and Machine Learning in Regulatory Submissions for Drug Development From 2016 to 2021. Clin Pharma and Therapeutics. 2022 Jun 16; cpt.2668. https://doi.org/10.1002/cpt.2668

52. Smith JA, Abhari RE, Hussain Z, Heneghan C, Collins GS, Carr AJ. Industry ties and evidence in public comments on the FDA framework for modifications to artificial intelligence/machine learning-based medical devices: a cross sectional study. BMJ Open. 2020 Oct; 10(10):e039969. https://doi.org/10.1136/bmjopen-2020-039969

53. Muehlematter UJ, Daniore P, Vokinger KN. Approval of artificial intelligence and machine learning-based medical devices in the USA and Europe (2015-20): a comparative analysis. The Lancet Digital Health. 2021 Mar; 3(3):e195-203. https://doi.org/10.1016/S2589-7500(20)30292-2

54. Ball R, Dal Pan G. "Artificial Intelligence" for Pharmacovigilance: Ready for Prime Time? Drug Saf. 2022 May; 45(5):429-38. https://doi.org/10.1007/s40264-022-01157-4

55. Chen Z, Liu X, Hogan W, Shenkman E, Bian J. Applications of artificial intelligence in drug development using real-world data. Drug Discovery Today. 2021 May; 26(5):1256- 64. https://doi.org/10.1016/j.drudis.2020.12.013