AI: A new lab partner in biotech

Politico

30-05-2025

THE REGULATORS
Biotech companies are using a new set of artificial intelligence products to hasten drug discovery and other scientific research, and they're energized by the Food and Drug Administration's apparent embrace of generative AI tools, said biotech leaders at a virtual Google Cloud AI roundtable Thursday.
What they said: Generative AI tools can absorb some tedious work that scientists would normally perform, like extracting novel insights from large datasets, including scientific literature or the human genome, the latter of which could take 'thousands of years' to comb through manually, said Ben Mabey, chief technology officer at Recursion, a pharmaceutical company.
'We can have [generative AI] agents do that and save the really hard problems for humans,' he said.
The remarks come as the FDA looks to universalize AI use in-house. Earlier this month, FDA Commissioner Marty Makary mandated the agency's divisions to use AI to conduct scientific reviews by the end of June.
Mabey and other biotech executives, including Girish Naganathan, chief technology officer of glucose monitoring tech company Dexcom, said that's a good sign for health AI's future as the FDA looks to build a regulatory framework for the technology. Now, companies like Dexcom are 'engaged in providing feedback to and seeking input from the FDA as we take steps to incorporate gen-AI into our products,' Naganathan said.
Artificial intelligence could help the FDA achieve other goals it set earlier this year, particularly one that aims to phase out animal testing with 'more effective, human-relevant methods,' Mabey said. 'We can replace [animals] with [AI] models that are more predictive, not just of the animals but of the humans we care about,' he said. 'It's a great North Star to work towards.'
Why it matters: Biotech companies are sizing up the FDA's relationship with AI to understand how the agency might regulate those tools. Regulations will likely influence their bottom line and dictate, for example, how many hurdles a company must clear to bring a drug borne out of AI to market.
WELCOME TO FUTURE PULSE
This is where we explore the ideas and innovators shaping health care.
Scientists have developed gene delivery 'trucks' that can target specific brain cells, which could lead to therapies for neurodegenerative diseases like Parkinson's and Alzheimer's, The Washington Post reports.
Radiologists are skeptical about how well AI can detect and diagnose cancer, even as the Food and Drug Administration clears the technology to analyze mammograms and imaging centers adopt these programs, STAT reports.
Share any thoughts, news, tips and feedback with Danny Nguyen at [e-mail], Carmen Paun at [e-mail], Ruth Reader at [e-mail], or Erin Schumaker at [e-mail].
Want to share a tip securely? Message us on Signal: Dannyn516.70, CarmenP.82, RuthReader.02 or ErinSchumaker.01.
Researchers at the National Institutes of Health used machine learning to identify an early biomarker of aggressive breast cancer and help predict which patients could be at high risk of recurrence or death from the disease.
The findings: The researchers used machine learning to detect minute changes in more than 9,000 samples of connective breast tissue, called stromal tissue, and found that significant abnormalities were associated with more aggressive forms of breast cancer and greater mortality, particularly in women with invasive forms of the disease.
An analysis of the stromal tissue samples — from healthy women, women with benign breast disease and women diagnosed with invasive breast cancer — suggested that so-called disruptive tissue might be tied to common risk factors for aggressive breast cancer, such as being Black, obese or young.
The study also found that significantly abnormal tissue was linked to higher risks of aggressive cancer and faster progression from benign breast disease to cancer than tissue with little to no stromal disruption.
Those disruptions can be influenced by chronic inflammation and wound healing, but according to Dr. Mustapha Abubakar, one of the study authors, 'We don't know which one comes first: inflammation [and wound healing] or stromal disruptions.' It's also not yet clear how people can prevent stromal anomalies from occurring.
Why it matters: The findings come as breast cancer incidence is rising in young women. A 2024 study by the American Cancer Society found that breast cancer incidence among women under 50 was increasing at 1.4 percent a year — about twice as fast compared with women older than 50. The disease is the second-leading cause of cancer death in women — only behind skin cancers.
These findings, if replicable and transferable, could offer scientists a therapeutic target. It also gives women another reason to get biopsies, Abubakar said; stromal disruptions don't manifest in anything we can see with our naked eye and can only be detected under a microscope.
It could take years for these findings to yield tangible impacts for women in clinical settings, Abubakar added. But scientists are moving faster than ever by using machine learning to identify patterns or anomalies in images of tissue, sequences of genomes and other data sets.