The use of artificial intelligence (AI) is growing, from banking and biodiversity to travel and transport. One of the areas where its presence is significant is healthcare, and its existing and potential applications include: [1-6]
One of the key areas where AI is playing a significant role is drug discovery and development. Currently, drug discovery can involve trial-and-error experimentation to select compounds, followed by large-scale testing. This hit and miss process can be costly and lengthy – up to 15 to 20 years of research – and may have low levels of accuracy and high failure rates. It also doesn’t always predict the efficacy and safety that the candidate compounds will exhibit in clinical trials. By analysing large quantities of data from benchtop research, preclinical trials and patient records, AI has potential to make major improvements to this route to the clinic through the identification of new targets and prediction of efficacy, safety and toxicity. [7, 8]
A (very) brief introduction to AI
In AI, computers and machines can simulate human comprehension, learning, creativity and autonomy, and mimic problem solving and decision making. AI is built on machine learning (ML – systems that learn from historical data) and deep learning (machine learning models that mimic human brain function). Generative AI (GenAI) builds further on machine learning and deep learning to create original text, images, video and other content. [9]
Identifying drug targets
Machine learning algorithms can be used to analyse large datasets and identify the targets that are mostly likely to interact with a potential drug. The datasets used include gene expression profiles, protein-protein interaction networks, biological pathways, clinical and chemical databases and unstructured data such as scientific literature. The targets can then be prioritised using ML algorithms such as support vector machines (SVMs) and neural networks. [6, 10]
ChatGPT (a natural language processing system) and large language models (LLM) are also being used to identify new drug targets. [8]
Finding better drugs
Ideal drugs will have on-target effectiveness with few or no off-target effects. Deep learning trained with known compounds and properties can be used to find existing molecules or design new molecules that have the desired activity, solubility and safety. Virtual screening, which can be structure-based or ligand-based, can speed the selection of candidate drugs. [3, 6, 11]
The combination of deep learning and interpretable machine learning can support de novo drug design and molecular dynamics in drug discovery. The parameters used include molecular similarity, quantitative structure–activity relationship (QSAR) and the process of generation of molecules. [3, 11]
New drugs can also be discovered using AI-powered imaging technology. Microscopy images from human cell assays, taken when cells have genes knocked out or different compounds added, can be compared using trained AI models that indicate biological differences between samples, thus identifying the compounds that could be the most effective. [10]
Predicting efficacy and safety
There are challenges with using animal studies and clinical trials to determine efficacy, safety and toxicity, as animal studies do not always fully predict either safety or efficacy and clinical trials are expensive and lengthy, and may put people at risk.
As an example, in 2006, six healthy volunteers were given TeGenero’s CD28 superagonist antibody TGN1412 in a Phase I clinical trial. The drug was in development for B-cell disorders and autoimmune disease, and animal studies had shown no significant drug-related adverse effects. The volunteers were given a dose 500 times smaller than that shown to be safe in animal studies. After receiving a single infusion of the drug at Northwick Hospital in London, UK, all six volunteers experienced organ failure. They were admitted to intensive care. One volunteer lost fingers and toes. While all survived, they may be at risk of future cancers or autoimmune diseases. [12, 13]
AI could better predict dangerous adverse effects by analysing large quantities of medical and scientific data. Machine learning and deep learning, trained with data on toxic and non-toxic drug compounds, including biological and clinical activity from preclinical and clinical trials, can predict the activity of new compounds. [6, 7]
ChatGPT and LLM can provide information on pharmacokinetics, pharmacodynamics and toxicity to support drug selection. [8]
Drug-drug interactions can result in increased levels of adverse effects or in increased/reduced efficacy. This is a particular issue in people with chronic disease or older people who are more likely to be taking a combination of drugs. By analysing data on drug interactions, AI, including ChatGPT and LLM, can help researchers to select potential leads with lower risk of drug-drug interactions, and support physicians in selecting the best drug regimes for individuals. [7, 8]
By improving all of these steps in drug discovery, AI has potential to speed up drug development by failing unsuitable compounds fast, thereby reducing the currently very poor success rate of drugs in development; only around 10% of drug candidates make it further than Phase I clinical trials. [6, 10]
The challenges and benefits
As well as benefits, AI comes with challenges. These include: [5, 7, 14]
Patient privacy
Ethical challenges, such as fairness and bias
Inability to replace the expertise and experience of human researchers
Need for validation by human researchers
Drug regulators’ lack of familiarity with AI
By working together, researchers, regulators, physicians and patients can overcome these challenges.
The use of AI has already been validated. As of 2023, there were more than 150 small molecule drugs in the discovery phase and more than 15 in clinical trials with biopharma companies that were developed using an AI-first approach. [15]
In an analysis of over 100,000 investigational and approved drugs in a global research and development database, as of 11 February 2024, 164 investigational drugs and one approved drug had been developed using AI, with 28% using machine learning (ML) and 17% using deep learning. [5, 8]
It’s unlikely that AI will ever replace human researchers. However, combining AI techniques with human expertise and experience could help make drug discovery more efficient, accurate, faster and more cost effective. [7, 14]
References
1. No longer science fiction, AI and robotics are transforming healthcare. Last accessed: 20 November 2024. Available from: https://www.pwc.com/gx/en/industries/healthcare/publications/ai-robotics-new-health/transforming-healthcare.html.
2. Mundell, I. AI for healthcare. Imperial Enterprise. Last accessed: 20 November 2024. Available from: https://www.imperial.ac.uk/stories/healthcare-ai/.
3. Bajwa, J., et al., Artificial intelligence in healthcare: transforming the practice of medicine. Future Healthc J, 2021. 8(2): p. e188-e194.
4. Ferry, S. and C. Amadi-Livingstone. AI in healthcare: navigating the noise. NHS Publisher. Last accessed: 20 September 2024. Available from: https://www.nhsconfed.org/publications/ai-healthcare.
5. Druedahl, L.C., et al., Use of Artificial Intelligence in Drug Development. JAMA Netw Open, 2024. 7(5): p. e2414139.
6. Rehman, A.U., et al., Role of Artificial Intelligence in Revolutionizing Drug Discovery. Fundamental Research, 2024: p. 2024.04.021.
7. Blanco-Gonzalez, A., et al., The Role of AI in Drug Discovery: Challenges, Opportunities, and Strategies. Pharmaceuticals (Basel), 2023. 16(6).
8. Pal, S., et al., ChatGPT or LLM in next-generation drug discovery and development: pharmaceutical and biotechnology companies can make use of the artificial intelligence-based device for a faster way of drug discovery and development. Int J Surg, 2023. 109(12): p. 4382-4384.
9. Stryker, C. and E. Kavlakoglu. What is AI? . Last accessed: 16 August 2024. Available from: https://www.ibm.com/topics/artificial-intelligence.
10. Brazil, R., How AI is transforming drug discovery. The Pharmaceutical Journal, 3 July 2024. Available from: https://pharmaceutical-journal.com/article/feature/how-ai-is-transforming-drug-discovery.
11. Bai, Q., et al., Application advances of deep learning methods for de novo drug design and molecular dynamics simulation. WIREs Computational Molecular Science, 2022. 12: p. e1581.
12. Attarwala, H., TGN1412: From Discovery to Disaster. J Young Pharm, 2010. 2(3): p. 332-6.
13. Roxby, P., Northwick Park drug trial disaster - could it happen again? BBC News, 24 May 2013. Available from: https://bbc.co.uk/news/health-22556736.
14. Chakraborty, S., et al., Artificial intelligence (AI) is paving the way for a critical role in drug discovery, drug design, and studying drug–drug interactions – correspondence. Int J Surg., 2023. 109(10): p. 3242-3244.
15. Liu, Y., Y. Chen, and L. Han, Bioinformatics: Advancing biomedical discovery and innovation in the era of big data and artificial intelligence. The Innovation Medicine, 2023. 1(1): p. 100012.
About the Author:
Suzanne Elvidge
Based in the north of England, Suzanne Elvidge is a freelance medical writer with a 30-year experience in journalism, feature writing, publishing, communications and PR. She has written features and news for a range of publications, including BioPharma Dive, Pharmaceutical Journal, Nature Biotechnology, Nature BioPharma Dealmakers, Nature InsideView and other Nature publications, to name just a few. She has also written in-depth reports and ebooks on a range of industry and disease topics for FirstWord, PharmaSources, and FierceMarkets. Suzanne became a freelancer in 2006, and she writes about pharmaceuticals, consumer healthcare and medicine, and the healthcare, pharmaceutical and biotechnology industries, for industry, science, healthcare professional and patient audiences.
ALL
Pharma in China
Pharma Experts
Market News
Products Guide
Brand Story
Pharma Sources Insight January 2025
