Introduction
The potential for artificial intelligence (AI) and machine learning (ML) applications in the scope of the drug discovery market reached 1.1 billion in 2022, with expectations to reach 4 billion by 2027.[8] The exponential growth of AI/ML reflects its impact on the healthcare system. AI/ML technologies are redefining the traditional approaches toward biomedical product development by simplifying many drug development and repurposing processes. The byproduct of an AI-fueled ecosystem is that it benefits patients by allowing new drug offerings to reach a broader patient population more quickly and with greater accessibility.[15]
The applications of AI/ML now cover critical facets of drug development, including predictive and statistical modeling, data mining and genomic and proteomic-specific data library creations, patient recruitment drug screening, pharmaceutical manufacturing, and much more.[9] For example, by combining the power of efficient computing with large scientific datasets, AI/ML can deliver statistical and predictive models of various drug compounds, thereby augmenting precise decision-making power in an expeditious manner.[2]
Having pushed through impressive milestones, AI-enabled technology is now a fully realized research tool with enormous potential. Earlier in 2023, a pharmaceutical company submitted to the FDA for Phase II clinical trials a first-ever orphan drug candidate solely designed using AI. Impressively, the AI-developed drug molecule required significantly fewer resources to progress through preclinical development compared to human processes requiring approximately 30 months from discovery to Phase 1 trials.[10]
Drug Discovery and Development
Drug Target, Identification, and Prioritization
Determining a viable biological target to develop a drug candidate is accomplished by sifting through copious amounts of information from a vast array of resources. The intent is to correlate the results of various disease characteristics and how they relate to other bits of related data. Using sophisticated algorithms, AI/ML-enabled tools can mine large sets of validated data to provide the information necessary to identify biological targets and prioritize them according to viability. Utilizing AI/ML in this capacity enables drug targeting to increase the overall efficiency of this phase discovery, streamlining a normal 21-day process to approximately three days.[2]
Preclinical Research
As an industry, preclinical research for drug development is expensive and timely. Some estimates indicate that it takes $2.3 billion to develop a drug from initial discovery to market.[3] Preclinical research using AI/ML methodologies assists in reducing both the timelines and the cost structures. They also reduce the failure rate because they can effectively predict molecular compatibility.
Biological Targeting
AI increases the ability to interpret biomedical information. By integrating what is known about a disease and its genetic predispositions, researchers can gain greater insight into these relationships between the two and construct more accurate biological targeting strategies. AI combines all the varying facets of drug discovery research (such as chemistry, statistics, pathology, biochemistry, physiology, and therapeutics) to garner greater experimental data and optimize drug development.
Screening New Molecular Entities
The number of molecular entities approved by regulatory bodies has increased dramatically in recent years. Importantly, the contributory data acquired by AI/ML bridges the gaps between translational and molecular drug interactions to generate accurate data on:
- Identifying unknown biomarkers
- Comparing target engagement and binding sites
- Determining heterogenic patterns
- Elucidating mechanisms of action and validation
By providing detailed information on these factors, AI/ML can increase the efficiency and accuracy of preclinical research.
AI and ML improve the variable techniques for screening new molecular entities, including ligand-based virtual screening, AI-augmented virtual screening, and structural 3D-based screening. These screening methods help identify hits and protein binding pockets needed for enhanced molecular clarity, sequencing data, and off-target possibilities.[19]
The beauty of AI/ML is that it can help elucidate the structural components of natural and synthetic molecules and compounds to uncover their bioactivity and potential for use. More importantly, AI/ML algorithms are so accurate that they reduce modeling errors and can classify problems with linear drug relationships. In this manner, the AI-enabled design assists in modeling potential applications while assessing and categorizing the qualitative values related to bioactive domains for future use. Additionally, this strategy can facilitate the repurposing of existing drugs and discover if they harbor anti-inflammatory, antimicrobial, and anti-cancer properties.[22]
Molecule Design and Synthesis Planning
Chemically speaking, AI/ML reduces the effort required to generate, discover, and develop a new drug. Computer-aided synthesis planning also facilitates future processes by creating chemical libraries that contain known best practices relating to formulas, techniques, and proven organic-chemistry template extraction methods. AI/ML help to overcome the known challenges in this chemical space related to costs and scale-ability.[5]
Structural Targeting and Binding Site Applications
Preclinical drug targeting hinges on pinpointing viable pharmaceutical components within protein families. However, researchers find that some components are not adequately represented in the contents of the global Protein Data Bank.[7] Structural information provides a clear window to identify areas of resistance potential, selectivity drivers, unknown mechanisms of action, and comprehensive ligand assessment. These factors correspond to drug development by offering insight into which sites will bind effectively to molecular components. AI substitutes the use of homologous modeling because of the comprehensive strength of data gained in the process.[7]
Interpreting Non-clinical Data
Non-clinical Research
Non-clinical research assists the drug development process by organizing collective pharmacokinetic (PK) and pharmacodynamic (PD) information acquired from animal and non-animal studies. The analysis of these pharmacometrics constitutes the means of incorporating non-clinical data into predictive models for clinical testing.[6] The computing power of AI/ML can leverage these models to find appropriate dosage regimens and further the scope of physiological-based PK and PD complementary data acquisition.[6] Gains in understanding drug discovery and development are rapidly emerging because of the creation of deep neural networks made possible by AI/ML technologies.[12; 14]
Clinical Research
Through the lens of expert clinical trial teams, AI/ML boasts considerable advantages in many critical areas of patient recruitment, selection, and management. Patient recruitment is fundamental in getting novel drug candidates through development and into the market. Recruitment can also be a critical means for patients who otherwise may miss the opportunity to participate in developing lifesaving interventions. The adoption of AI-enabled data mining helps identify study candidates based on patient electronic health records (EHRs). Moreover, data mining of EHRs can identify study subjects in the safest way possible. This is especially important for studies involving special-case or vulnerable subject populations (such as pregnant, elderly, pediatric, and rare-disease populations).[4]
AI/ML provides a mechanism to tackle traditional challenges such as compliance and retention by enhancing communication channels and monitoring via smartphone devices, wearables, and implanted devices. While automation may appear too technical and impersonal of an approach to patients, the FDA has identified improvements in safety profiles, adverse reporting, and risk mitigation as goals to improve the transparency of clinical trial information.[21]
AI/ML also has implications for assessing a site for a clinical study. Its algorithmic tools can assist in site selection through its ability to appraise study sites, their past performance, and future recruiting potential, with greater detail and clarity.[11]
Integrating novel AI tools such as natural language processing (NLP) can give investigators and sponsors the insight needed to minimize potential pitfalls and refine inclusion/exclusion criteria to optimize a study population. Study simulation prior to a clinical trial is also a tool that aids investigators in the planning phases of study design, thereby yielding protocols that may likely produce better results.[20]
Post-market Surveillance and Approval
Within the last few years, the FDA has received a growing number of submissions that specify AI and ML spanning multiple interdisciplinary sectors of drug design (more than 100 in 2021).[15] Consequently, the FDA is homing in on avenues to regulate AI and ML to facilitate its safe and effective use. The FDA’s Sentinel Initiative, the Center for Biologics Evaluation and Research (CBER) Biologics and Safety system, and the Center for Devices and Radiological Health’s National Evaluation System for Health Technology are all using AI and ML techniques to generate and evaluate post-market safety data. While these efforts continue, the FDA is also seeing to identify best practices for AI and ML across the preclinical, premarket, and post-market phases of a drug, biologic, or device.[15; 16]
ISTAND: A Pilot Program Initiated by the FDA
The FDA Center for Drug Evaluation and Research created its Innovative Science and Technology for New Drugs (ISTAND) to nurture the adoption of proven drug development tools. As such, it includes a submission allowance that references AI and ML tools. The pilot program partners the FDA with sponsors to assess AI tools capable of aiding in clinical study design and conduct with the goal of promoting the use of AI and its incorporation into a developing regulatory framework.[15; 17]
Adverse Case Reporting
Pharmacovigilance (PV) is one of the more critical components of the drug development process. PV relies on adverse event reports from patients and healthcare providers. However, drug manufacturers have struggled to gather and interpret data as there is often a lag time and incompleteness of reports from patients and physicians. AI and ML offer innovative means of data extraction from individual case reports identifying, in near real-time, when a post-market adverse event occurs and the causal relationship of the drug to the event, thus greatly helping case processing.[13] AI and ML can tap easily into source data, easing the adverse reporting verification process. In short, AI tools hold great potential to extract information in such a way that highlights the accuracy of case reporting and validation.[13]
Advanced Pharmaceutical Manufacturing
Industry 4.0 is a term used to represent the fourth tier of the industrial revolution. The future of manufacturing lies in a revolutionary movement to streamline and strengthen manufacturing activities. The technological advances spurred by AI and ML are aiming towards a collective hub of sorts to increase the level of control throughout the value chain and life cycle of a manufactured product.[18] With global connectivity and digitization, pharmaceutical manufacturers can bring Industry 4.0 principles to bear on manufacturing practices that can achieve a high level of productivity and a low level of wasted resources.[1]
Manufacturing Process Optimization Strategies
Many manufacturing improvements fueled by AI and ML are seen in monitoring capabilities. Smart monitoring allows for real-time assessment of equipment, controls, packaging, and labeling accomplished in part by augmented visual surveillance technology. This technology usage translates into minimizing errors because it can point out interruptions and deviations associated with the manufacturing process.[15] Trend monitoring proactively tackles procedural problems such as disruptions related to volume, packaging, and equipment failures. The methodology is designed to deliver metrics that allow for continual improvements in manufacturing speed, accuracy, and reliability.[15]
FDA Response to AI and ML
Creating a colossal wave of change within the structure of the existing drug discovery process, AI and ML are steadily influencing amendments within the current regulatory framework. Evolving quickly into mainstream use, the FDA seeks to create guidance on how AI can be integrated into the different aspects of clinical development.[15]
Understanding that the AI landscape is rich with innovations and optimization strategies, the FDA has highlighted its value and uses and noted AI’s limitations and challenges. To achieve greater clarity, the FDA has encouraged feedback from outside stakeholders.[15] It has identified three areas of concern:
- human-led governance, accountability, and transparency.
- quality reliability and representativeness of data; and
- module development, performance, monitoring, and validation
Although the FDA has issued neither regulations nor guidance on AI and ML, we can expect the agency to grapple with the issues raised by these tools for years to come. Their efforts will be needed to engage with AI technologies in a way that keeps subjects safe while improving trial objectives.
Summary
Artificial intelligence and machine learning are poised to overhaul biomedical product research and development radically. They impact the process from early discovery through post-approval surveillance with the potential of improving efficiencies and providing cures and diagnostics that otherwise wouldn’t be possible.
References
[1] Arden, N. Sarah, Adam C. Fisher, Katherine Tyner, Lawrence X. Yu, Sau L. Lee, and Michael Kopcha. 2021. “Industry 4.0 for Pharmaceutical Manufacturing: Preparing for the Smart Factories of the Future.” International Journal of Pharmaceutics 602:120554.
[2] Benjamens, Stan, Pranavisingh Dhunnoo, and Bertalan Meskó. 2020. “The state of artificial intelligence-based FDA-approved medical devices and algorithms: an online database.” npj Digital Medicine 3(118).
[3] Deloitte. 2023. “Seize the digital momentum: Measuring the Return from Pharmaceutical Innovation 2022.” Accessed June 30, 2023.
[4] Glicksberg, Benjamin S. Glicksberg, Riccardo Miotto, Kipp W. Johnson, Khader Shameer, Li Li, Rong Chen, and Joel T. Dudley. 2018. “Automated Disease Cohort Selection Using Word Embeddings from Electronic Health Records.” Pacific Symposium on Biocomputing 2018 23:145-56.
[5] Ivanekov, Yan A., Daniil Polykovskiy, Dmitry Bezrukov, Bogdan Zagribelnyy, Vladimir Aladinskiy, Petrina Kamya, Alex Aliper, Feng Ren, and Alex Zhavoronkov. 2023. “Chemistry42: An AI-Driven Platform for Molecular Design and Optimization.” Journal of Chemical Information and Modeling 63(3):695-701.
[6] Keutzer, Lina, Huifang You, Ali Farnoud, Joakim Nyberg, Sebastian G. Wicha, Gareth Maher-Edwards, Georgios Vlasakakis, Gita Khalili Moghaddam, Elin M. Svensson, Michael P. Menden, and Ulrika S. H. Simonsson. 2022. “Machine Learning and Pharmacometrics for Prediction of Pharmacokinetic Data: Differences, Similarities and Challenges Illustrated with Rifampicin.” Phamaceutics 14(8):1530.
[7] Kozlovskii, Igor, and Petr Popov. 2021. “Structure-Based Deep Learning for Binding Site Detection in Nucleic Acid Macromolecules.” NAR Genomics & Bioinformatics 3(4):lqab111.
[8] MarketsandMarkets Research Pvt. Ltd. 2023. “AI in Drug Discovery Market is Expected to Reach $4 Billion.” Global Newswire, May 29.
[9] Paul, Debleena, Gaurav Sanap, Snehal Shenoy, Dnyaneshwar Kalyane, Kiran Kalia, and Rakesh K. Tekade. 2021. “Artificial Intelligence in Drug Discovery and Development.” Drug Discovery Today 26(1):80-93.
[10] Philippidis, Alex. 2023. “Insilico Gains FDA’s First Orphan Drug Designation for AI Candidate.” Genetic Engineering & Biotechnology News, February 10.
[11] Reed, Jane. 2022. “How AI-Based Technologies Improve Clinical Trial Design, Site Selection and Competitive Intelligence.” Drug Discovery & Development, July 21.
[12] Samek, Wojciech, Grégoire Montavon, Sebastian Lapuschkin, Christopher J. Anders, and Klaus-Robert Müller. 2021. “Explaining Deep Neural Networks and Beyond: A Review of Methods and Applications.” Proceedings of the IEEE 109(3):247-78.
[13] Schmider, Juergen, Krishan Kumar, Chantal LaForest, Brian Swankoski, Karen Naim, and Patrick M. Caubel. 2019. “Innovation in Pharmacovigilance: Use of Artificial Intelligence in Adverse Event Case Processing.” Clinical Pharmacology & Therapeutics 105(4):954-61.
[14] Subbiah, Vivek. 2023. “The Next Generation of Evidence-Based Medicine.” Nature Medicine 29:49-58.
[15] U.S. Food and Drug Administration (FDA). n.d. “Using Artificial Intelligence & Machine Learning in the Development of Drug & Biological Products.” Accessed June 29, 2023.
[16] U.S. Food and Drug Administration (FDA). 2021. “Good Machine Learning Practice for Medical Device Development: Guiding Principles.” Accessed June 29, 2023.
[17] U.S. Food and Drug Administration. 2023. “Innovative Science and Technology Approaches for New Drugs (ISTAND) Pilot Program.” Accessed June 29, 2023.
[18] Vaidya, Saurabh, Prashant Ambad, and Santosh Bhosle. 2018. “Industry 4.0 – A Glimpse.” Procedia Manufacturing 20:233-8.
[19] Vijayan, R.S.K., Jan Kihlberg, Jason B. Cross, and Vasanthanathan Poongavanam. 2022. “Enhancing Preclinical Drug Discovery with Artificial Intelligence.” Drug Discovery Today 27(4) 967-84.
[20] Weissler, E. Hope, Tristan Naumann, Tomas Andersson, Rajesh Ranganath, Olivier Elemento, Yuan Luo, Daniel F. Freitag, James Benoit, Michael C. Hughes, Faisal Khan, Paul Slater, Khader Shameer, Matthew Roe, Emmette Hutchison, Scott H. Kollins, Uli Broedl, Zhaoling Meng, Jennifer L. Wong, Lesley Curtis, Erich Huang, and Marzyeh Ghassemi. 2021. “The role of machine learning in clinical research: transforming the future of evidence generation.” Trials 22(537)
[21] Young, Albert T., Dominic Amara, Abhishek Bhattacharya, and Maria L. Wei. 2021. “Patient and General Public Attitudes Towards Clinical Artificial Intelligence: A Mixed Methods Systematic Review.” The Lancet Digital Health 3(9):e599-611.
[22] Zhang, Ruihan, Xiaoli Li, Xingjie Zhang, Huayan Qin, and Weilie Xiao. 2020. “Machine Learning Approaches for Elucidating the Biological Effects of Natural Products.” Natural Product Reports 38(2):346-61.
