The FDA and artificial intelligence

The FDA and artificial intelligence

By Dr Nicola Davies 

The research and development of pharmaceuticals is a costly, lengthy, and high-risk venture, spread over four stages of studies: drug discovery, preclinical research, clinical research, and post-marketing studies on safety and effectiveness. The journey of a new drug to the market is estimated to take approximately 10 years and cost billions of US dollars, says Dr Nicola Davies in her exclusive article on the US Food and Drug Administration.1 

A modern approach for overcoming these challenges is the use of artificial intelligence (AI) in drug development. The coming of AI is a boon for the pharma industry because it can significantly cut the cost of development and time-to-market for new drugs. For example, in the race to develop vaccines and other therapeutics for COVID-19, AI-based virtual screening of both new and repurposed molecules can accelerate discovery.2

The FDA defines AI as “the science and engineering of making intelligent machines, especially intelligent computer programs.”3 In recent years, AI has been increasingly used in drug repurposing, which involves developing an already approved drug for a different indication or different population; optimizing and expediting patient recruitment in trials; and, post-approval safety surveillance.1 However, AI holds huge benefits not only for pharma but also for the FDA; AI can facilitate the agency’s regulatory activities related to drug development and approval. 

When it comes to digital technology, the FDA has been attempting to balance its role as an “innovation enabler” and a “safety watchdog.”4 We look at how the FDA has been encouraging AI in drug development, as well as implementing AI in its own processes.

The FDA’s support for AI in drug development

Since the 21st Century Cures Act of 2016, the FDA has been supporting the use of innovative approaches to drug development and regulatory processes, such as the acceptance of real-world evidence and real-world data for making regulatory decisions.5

AI in drug repositioning

Dr Zhichao Liu, a senior technical leader in the AI Research Force (National Center for Toxicological Research) of the FDA, developed the standard pipeline using AI to facilitate drug repositioning efforts by the industry and accelerate drug development from a regulatory perspective.6 Additionally, he led efforts in developing AI/machine learning solutions in predictive toxicology, a rising field in toxicology that allows the prediction of toxicity in humans so that the use of animals for safety evaluation can be avoided or reduced. Some of these models have been adopted within industry and regulatory frameworks.

In addition, the FDA is currently developing a AI component for its FDALabel database, which provides easy and quick access to the labels of approved drugs, including data on chemical structure, pharmacological effects, drug–drug interactions, and toxicity; these data can be used for identifying potential drug repurposing strategies.7 AI is expected to enable the use of “customized and fine-tuned public language models and algorithms” when entering queries; this will improve the specificity and relevance of the query results retrieved and facilitate efforts toward drug repurposing.

AI in adverse event monitoring

Patient narratives from clinical study reports and electronic health records (EHRs) have facilitated safety monitoring of marketed therapeutics; however, manual review of these data sources can be time-consuming and difficult. AI can be deployed to further improve the efficiency, quality, and consistency of this critical task, while reducing reporting burden.1 

One such initiative is underway by the FDAs Center for Drug Evaluation and Research, which is developing a Deep Learning MedDRA encoder (MedDRA-DeepCoder) to efficiently identify and analyze adverse events reported in clinical studies. Narratives from the FDA Adverse Event Reporting System (FAERS) were successfully used to train and validate models for this purpose, using deep learning or machine learning methodologies.6

In addition, the Center for Biologics Evaluation and Research of the FDA has launched the Biologics Effectiveness and Safety Innovative Methods (BEST IM) initiative, which utilizes AI, machine learning, natural language processing, and other innovative technologies to improve adverse event reporting for marketed biological products from EHRs.6 A successful application of this platform is the BEST chart review tool, which extracts data on potential cases flagged by algorithms and presents them for clinical review. Subsequently, the tool pre-populates an individual case safety report, which is reviewed for submission to the FDA.

AI in the drug review process

AI can improve the efficiency and quality of regulatory assessments, such as the review of drug approval submissions. The Office of Generic Drugs (OGD) successfully tested its in-house-developed data/text analytics tool, Bioequivalence Assessment Mate (BEAM), for improving the efficiency and consistency of bioequivalence assessment for generic drug approvals. This is a significant development because it can enable the agency to meet assessment timelines despite the high number of abbreviated new drug application (ANDA) submissions.6

The FDA’s role in the future of AI for drug development

The FDA has been supporting research on the use of AI and the development of algorithms for critical aspects of biological and pharmaceutical research, such as safety, quantitative structure–activity relationship (QSAR) modeling, and genomics studies.6

Additionally, the FDA indirectly supports AI in drug development as part of the Tox21 Consortium; this group aims to develop rapid and efficient techniques for analyzing the safety of various substances, including medical products.8 The Consortium allows access to its datasets containing toxicity data and assay information for various substances. The vast amount of toxicological data available in Tox21 datasets enables research and development of machine learning based models for predictive toxicology.6

Dr Liu has proposed a DeepFake model framework for drug repurposing against COVID-19, based on precision medicine. The model is designed for understanding the interaction between the immune system and mitochondria to determine associations between this data and the severity and pre-existing conditions of patients with COVID-19. Next, FDA-approved drugs and drug candidates under investigation are mapped to this data, which yields a list of potential candidates for repurposing in patients with COVID-19 presenting with different manifestations.6

Considering that the role of AI in drug development has been strengthening, the FDA needs to integrate tech companies, including tech giants such as Apple, into its existing group of stakeholders.4 However, in doing so, the agency should ensure that technological innovation does not compromise the safety and effectiveness of AI solutions.

References

  1. Chen Z, Liu X, Hogan W, Shenkman E, Bian J. Applications of artificial intelligence in drug development using real-world data. Drug Discov Today. 2020;S1359-6446(20)30531-6. doi: 10.1016/j.drudis.2020.12.013
  2. Singh N, Villoutreix BO. Resources and computational strategies to advance small molecule SARS-CoV-2 discovery: Lessons from the pandemic and preparing for future health crises. Comput Struct Biotechnol J. 2021;19:2537-2548. doi: 10.1016/j.csbj.2021.04.059.
  3. US FDA, 2021. Artificial Intelligence and Machine Learning in Software as a Medical Device. [Online]

Available at: https://www.fda.gov/medical-devices/software-medical-device-samd/artificial-intelligence-and-machine-learning-software-medical-device [Accessed May 4, 2021]

  1. Lievevrouw E, Marelli L, Van Hoyweghen I. The FDA's standard-making process for medical digital health technologies: co-producing technological and organizational innovation. Biosocieties. 2021:1-28. doi: 10.1057/s41292-021-00232-w
  2. US FDA, 2020. Real-World Evidence. [Online]

Available at: https://www.fda.gov/science-research/science-and-research-special-topics/real-world-evidence [Accessed May 4, 2021]

  1. US FDA, 2021. 2021 FDA Science Forum. [Online]

Available at: https://www.fda.gov/media/149502/download [Accessed May 3, 2021]

  1. Fang H, Harris S, Liu Z, Thakkar S, Yang J, Ingle T, Xu J, Lesko L, Rosario L, Tong W. FDALabel for drug repurposing studies and beyond. Nat Biotechnol. 2020;38(12):1378-1379. doi: 10.1038/s41587-020-00751-0
  2. Tox21, 2021. Toxicology in the 21st Century. [Online]

Available at: https://tox21.gov/ [Accessed May 3, 2021]

要查看或添加评论,请登录

社区洞察

其他会员也浏览了