| | | | |  | | By Erin Schumaker, Daniel Payne and Ruth Reader | | | | | 
AI's already improving cancer diagnoses. | Getty Images | Artificial intelligence capable of diagnosing cancer is one of the most compelling health care use cases for the technology. Researchers are making promising strides toward that goal, writes National Institutes of Health Director Monica Bertagnolli in a new director’s blog entry. How so? Bertagnolli points to an effort at Harvard Medical School to create a model that can diagnose multiple types of cancer, an advance beyond existing models that focus on just one type. The Harvard researchers have done it by training their system to read digital slides of tumor tissue and to analyze the tissue surrounding the tumor to detect cancer and predict how a patient might respond to treatment. The model the researchers have developed, the Clinical Histopathology Imaging Evaluation Foundation, or CHIEF, has trained on more than 15 million images. The researchers have also tested the model on 19 cancer types, using more than 19,400 whole-slide images from 32 datasets from 24 hospitals worldwide. CHIEF was up to 36 percent more accurate than other state-of-the-art AI methods in detecting cancer cells; identifying tumor origins; predicting patient outcomes; and identifying genes and DNA patterns linked to patients’ responses to surgery, chemotherapy, radiation and immunotherapy, according to the researchers. “If validated further and deployed widely, our approach, and approaches similar to ours, could identify early on cancer patients who may benefit from experimental treatments targeting certain molecular variations,” Dr. Kun-Hsing Yu, study coauthor and assistant professor of biomedical informatics at the Blavatnik Institute at Harvard Medical School, said in a statement. Yu’s NIH-backed research was published in the journal Nature in September. What’s next? Training the model on images of precancerous and noncancerous tissue, and on images of tissue from rare cancers — plus exposing the model to more data so it learns to better identify cancers of varying aggressiveness. Even so: From the NIH director, a dash of realism: “This is all good news, but there’s much more work ahead before an AI model like this could be used in the clinic,” Bertagnolli wrote.
| | | | 
Lisbon, Md. | Shawn Zeller | | This is where we explore the ideas and innovators shaping health care. Researchers have found a way to look under mice's skin. Putting the food dye that gives Cheetos their color on mice made their skin temporarily transparent, allowing researchers to examine their organs. Share any thoughts, news, tips and feedback with Carmen Paun at cpaun@politico.com, Daniel Payne at dpayne@politico.com, Ruth Reader at rreader@politico.com, or Erin Schumaker at eschumaker@politico.com. Send tips securely through SecureDrop, Signal, Telegram or WhatsApp.
| | |  The government is putting money behind the search for new antibiotics. | Shutterstock | The Advanced Research Projects Agency for Health is tapping artificial intelligence to fight a major global threat: antimicrobial resistance — when bacteria, viruses, fungi and parasites stop responding to antibiotics. "The rise of antibiotic resistance threatens to turn once-treatable infections into life threatening ones,” ARPA-H Director Renee Wegrzyn said in a statement. “But with AI, we can accelerate the discovery of new antibiotics to address this threat like never before." How so? The $27 million program, called Transforming Antibiotic R&D with Generative AI to stop Emerging Threats, or TARGET, will be led by Boston-based biotech company Phare Bio, along with the Collins Lab at the Massachusetts Institute of Technology and Harvard’s Wyss Institute. The researchers have three main targets: — Tap generative AI to increase the number of antibiotic candidates. The research team will use the Broad Institute’s Drug Repurposing Hub and the University of California, San Francisco’s ZINC15 library, which hold 107 million molecule candidates combined, to screen for antibiotic activity. They’ll also use generative AI to design candidates from scratch. — Use deep learning to develop screening tools. Using machine learning, the team will develop digital tools to assess the effectiveness of molecule candidates and test them to evaluate whether they’ll meet clinical and regulatory standards. — Validate promising antibiotics. The group aims to discover 15 promising leads for new antibiotics, which it says would help replenish the global antibiotic pipeline. Why it matters: More than 2.8 million antimicrobial-resistant infections occur in the U.S. annually, according to the Centers for Disease Control and Prevention, and bacterial infections are a leading cause of death globally. But the development of new antibiotics is slow since they require extensive manual screening and testing — and most fail.
| | | | 
Hassabis will split half of the $1 million chemistry prize with his Google colleague. | Getty Images | The Nobel Prize in chemistry went this year to three people who are at the forefront of the future of biotech: a biochemist who designs new proteins and two computer scientists who developed an artificial intelligence model to predict protein structure. In 2003, one of the winners, David Baker, designed a new protein that didn’t exist in nature. Computer software he developed has since developed proteins that bind to the synthetic opioid fentanyl, which could be used to detect the drug in the environment. His lab at the University of Washington has also come up with other proteins that can be used as medicines, vaccines and tiny sensors, the Royal Swedish Academy of Sciences, which awards the prize, said Wednesday. Baker shares the prize with Demis Hassabis, the CEO of Google’s AI company, DeepMind, and John Jumper, a senior research scientist at the same company, both based in London. Hassabis and Jumper launched their AI model AlphaFold2 in 2020. It has since predicted the structure of nearly all 200 million proteins ever identified, said the Royal Swedish Academy of Sciences. More than two million people from 190 countries have used the AI model. “Among a myriad of scientific applications, researchers can now better understand antibiotic resistance and create images of enzymes that can decompose plastic,” the academy said about the significance of the AI model’s use. Why it matters: Proteins are the basis of life. They control and drive chemical reactions and can also function as hormones, antibodies or the building blocks of different tissues, the academy said. The convergence of protein design and AI could give the world new drugs and vaccines. Even so: Experts in biotech and AI have also warned that this convergence could be used for malicious purposes and called on governments to institute guardrails on such research. | | | | Follow us on Twitter | | | | Follow us | | | | |
