Artificial intelligence in medicine

Artificial Intelligence (AI) is poised to rapidly become a central part of society. How could it not? AI chatbots like ChatGPT can answer questions, have detailed conversations, boost productivity, compose essays, write computer code, and summarize complex topics all within moments.¹ In fact, the potential impact is so large that in October 2023, less than one year after the release of ChatGPT, the Biden Administration issued a first-of-its-kind Executive Order aimed at both promoting the promise and managing the risks of AI. This type of oversight will be important because AI will likely have a major impact on medicine too.

What is artificial intelligence?
AI is a simulation of human intelligence. To get AI to use information as a human would, scientists “train the AI”. This means feeding the computer program huge amounts of data, asking it questions, evaluating the answers, and tweaking the code based on those answers. When the answers match what a human would reasonably do, the AI is given new information, and it can make predictions based on what it had previously learned. For example, ChatGPT was trained with natural language from massive quantities of data sourced from the internet. To generate responses to questions, it uses its database, how it was trained, the prompt, and the prior sequence of words to predict the next word in a sentence.

Emerging applications of AI in healthcare
AI can be trained using almost any kind of information, including diagnostic images, chemical structures, and sets of molecular data derived from patients, along with the corresponding outcomes. With these types of data, there are many exciting applications for AI in healthcare, but three stand out as likely to have a major impact.

Improving outcomes via early screening and accurate diagnoses
Early detection of disease is key to achieving the best outcomes. For example, a diagnosis of breast cancer in early stages greatly lowers the likelihood of death from that cancer,² and AI is already improving this. A group in Sweden found that AI-supported mammography reading could emphasize suspicious findings on the images, which aided in the detection of 20% more cancers than human eyes alone.³

AI can also help ensure the accuracy of a diagnosis. Sometimes, it is not known where a metastatic tumor originated, and this complicates therapy because the origin dictates specific treatments.⁴ Recently, two research groups at Harvard trained AIs on tumor images and molecular data from thousands of cancers. They used the AIs to predict the origin of a cancer when it was unknown, and they were able to classify nearly half of these tumors—a major advance for these patients.^5,6

Accelerating drug development
Traditional drug development involves identifying a target, synthesizing and screening tens of thousands of chemical compounds, testing them in pre-clinical models (such cells-in-a-dish and animal models), and then moving into clinical trials. For a single drug, this can cost hundreds of millions of dollars and take nearly a decade.⁷ AI has the potential to significantly accelerate this process. AI can process structures for millions of proteins and predict the shape of molecules most likely to alter function.^8,9 In this way, the pool of prospective drugs can quickly be narrowed, saving an immense amount of time and resources. For context, it has been estimated that the number of possible drug-like molecules is astoundingly high,¹⁰ far exceeding the number of stars in the universe.¹¹ Without aid from AI, humans are likely to miss many potential treatments.

Indeed, AI-assisted drug development is already occurring. Scientists at Insilico Medicine used AI to predict a structure that inhibits the protein DDR1, a promising target for fibrosis, which often has a poor prognosis. The compound was synthesized and validated in less than two months.¹² Since then, the drug has moved into clinical trials at record speed and achieved FDA orphan drug status earlier this year.

Improving personalized medicine
Personalized medicine is a term for tailoring treatments to individual patients rather than to what works for the general population. It is used for certain cancers to target characteristics based on molecular profile. As examples, breast cancer patients whose tumors have HER2 expression may receive trastuzumab,¹³ and lung cancer patients with an EGFR mutation may receive Osimertinib.¹⁴ The current paradigm is generally limited to matching single alterations to a drug, but AI can expand this substantially.¹⁵ By training AI on molecular profiling data, as well as historical patient records, AI can predict drug responses based on combinations of molecular profiling variables that humans may miss.¹⁶

At PHM, we believe AI has the potential to transform many aspects of healthcare in ways that may improve patient outcomes. In fact, in certain cases, we have partnered with select AI-focused companies to enhance personalized medicine and help deliver the best of what’s possible in medicine.

References

1. Taecharungroj, V. “What Can ChatGPT Do?” Analyzing Early Reactions to the Innovative AI Chatbot on Twitter. Big Data and Cognitive Computing 7.1, 35 (2023).
2. Guo, F., Kuo, Y. F., Shih, Y. C. T., Giordano, S. H. & Berenson, A. B. Trends in breast cancer mortality by stage at diagnosis among young women in the United States. Cancer 124, 3500-3509, doi:10.1002/cncr.31638 (2018).
3. Lang, K. et al. Artificial intelligence-supported screen reading versus standard double reading in the Mammography Screening with Artificial Intelligence trial (MASAI): a clinical safety analysis of a randomised, controlled, non-inferiority, single-blinded, screening accuracy study. Lancet Oncol 24, 936-944, doi:10.1016/S1470-2045(23)00298-X (2023).
4. Bianchi, J. J., Zhao, X., Mays, J. C. & Davoli, T. Not all cancers are created equal: Tissue specificity in cancer genes and pathways. Curr Opin Cell Biol 63, 135-143, doi:10.1016/j.ceb.2020.01.005 (2020).
5. Lu, M. Y. et al. AI-based pathology predicts origins for cancers of unknown primary. Nature 594, 106-110, doi:10.1038/s41586-021-03512-4 (2021).
6. Moon, I. et al. Machine learning for genetics-based classification and treatment response prediction in cancer of unknown primary. Nat Med, doi:10.1038/s41591-023-02482-6 (2023).
7. Kaitin, K. I. Deconstructing the drug development process: the new face of innovation. Clin Pharmacol Ther 87, 356-361, doi:10.1038/clpt.2009.293 (2010).
8. Jumper, J. et al. Highly accurate protein structure prediction with AlphaFold. Nature 596, 583-589, doi:10.1038/s41586-021-03819-2 (2021).
9. Schneider, G. Automating drug discovery. Nat Rev Drug Discov 17, 97-113, doi:10.1038/nrd.2017.232 (2018).
10. Reymond, J. L. The chemical space project. Acc Chem Res 48, 722-730, doi:10.1021/ar500432k (2015).
11. Manojlovic, L. M. Photometry-based estimation of the total number of stars in the Universe. Appl Opt 54, 6589-6591, doi:10.1364/AO.54.006589 (2015).
12. Zhavoronkov, A. et al. Deep learning enables rapid identification of potent DDR1 kinase inhibitors. Nat Biotechnol 37, 1038-1040, doi:10.1038/s41587-019-0224-x (2019).
13. Guo, L. et al. Breast cancer heterogeneity and its implication in personalized precision therapy. Exp Hematol Oncol 12, 3, doi:10.1186/s40164-022-00363-1 (2023).
14. Schuler, M. et al. Personalized Treatment for Patients With Lung Cancer. Dtsch Arztebl Int, doi:10.3238/arztebl.m2023.012 (2023).
15. Amisha, Malik, P., Pathania, M. & Rathaur, V. K. Overview of artificial intelligence in medicine. J Family Med Prim Care 8, 2328-2331, doi:10.4103/jfmpc.jfmpc_440_19 (2019).
16. Rezayi, S., S, R. N. K. & Saeedi, S. Effectiveness of Artificial Intelligence for Personalized Medicine in Neoplasms: A Systematic Review. Biomed Res Int 2022, 7842566, doi:10.1155/2022/7842566 (2022).