Out of every 10,000 to 15,000 new chemical compounds identified during drug discovery, just five will make it into human clinical trials. More than 92 percent of these drugs will fail in trials, usually because they were not sufficiently safe or effective.
On average, 50 novel medicines make it to market each year, each requiring a decade or longer of effort. The development of a single drug delivered to patients can for one-third of a scientist’s career, if they’re fortunate to find such success.
Artificial intelligence promises to make these numbers less daunting by generating hypotheses, analyzing massive sets of data and interpreting results far more quickly, expansively and in depth than is possible by the human brain alone.
AI accelerates everything. Critically, it can identify patterns and correlations that might elude the human eye. It integrates data from different sources faster and more efficiently. Case in point: Sanju Sinha, a computational biologist at Sanford Burnham Prebys, developed with colleagues at the National Cancer Institute an AI tool that analyzed data from thousands of cancer samples from people all over the world, looking for differences between healthy and cancer cells, then predicting response and resistance to specific treatments. In early testing with three different cancers, the AI tool, called PERCEPTION, proved remarkably predictive.
It took Sinha and colleagues two-and-a-half years to develop PERCEPTION and publish their findings. They’re now seeking to expand the work with additional studies. At 30 years old, Sinha probably hasn’t even completed the first third of his career. He’s got both the tools and time for greater success.
Sinha illustrates another important point: the need for multidisciplinary talent. Sinha was trained as a mathematician, not as a biologist. His particular expertise helps make it possible to incorporate AI into biomedical research.
That’s a paradigm shift. Traditionally, medical schools are organized into classic departments in medical specialties. They must evolve to welcome faculty with completely new skills that are disease-agnostic.
Most headlines regarding AI and human health, of course, involve its use by doctors and in the actual treatment of patients.
ChatGPT, an AI chatbot that uses machine learning to understand and generate human-like dialogue, has successfully ed the United States Medical Licensing Examination, which doctors must take before obtaining a license to practice. Google DeepMind has created an AI language model trained on existing medical Q&A datasets that can offer “safe and helpful answers” to questions posed by health care professionals and patients.
Some physicians now use AI to determine treatment protocols, clinical tools and appropriate drugs more efficiently. AI can document patient encounters in near-real time, improving documentation. Chatbots help patients find available doctors, schedule appointments and even answer some basic questions.
AI is also used to diagnose some conditions requiring visual comparisons, such as diabetic retinopathy, a form of vision loss, and brain scans to discern subtle signs of disease and dementia. These algorithms can detect and identify minute pathologies in seconds, far quicker than even the most experienced clinicians.
To be sure, a technological change as deep and sweeping as AI is never simple nor easy. Roughly 60 percent of Americans in a 2022 Pew Research Center survey said they would be uncomfortable with the idea of AI being used in their own health care; 38 percent believed using AI to diagnose disease and recommend treatments would lead to better health outcomes but 33 percent believed it would lead to worse outcomes.
They were more optimistic about the potential of AI to reduce medical errors and to lessen racial/ethnicity biases that adversely affect health care access and treatment.
Such skepticism of AI is good. The need to empirically and persuasively prove AI’s therapeutic value is absolute. The National Institutes of Health recognizes that safe and responsible use of AI is a work in progress.
AI will never replace the need and value of human physicians who bring, well, a humanity not embedded in any database. But AI represents the biggest and boldest advance yet in biomedical research and clinical medicine, on par with CRISPR genome editing, nanomedicine, robotics, imaging technologies, precision medicine and mRNA vaccines.
Indeed, AI makes it possible to speed those achievements to greater effect.
A note: This essay was written by a human being, not ChatGPT or any similar computer program. I’m happy to report that when ChatGPT was later asked for its thoughts on this topic, our observations were pretty much aligned.
Brenner is a physician-scientist and president and chief executive officer of Sanford Burnham Prebys and lives in La Jolla.
