The AI drug discovery story gets a reality check
Generative AI has become the pharmaceutical industry's favorite story about its own future — models that propose new molecules, predict how they behave, and compress years of lab work into months. Today's coverage focused on where that story actually holds up.
The clearest use case remains "hit identification," the earliest step in making a medicine. It's a needle-in-a-haystack problem: out of an almost unimaginable number of possible molecules, which few are worth testing in a lab? That's exactly the kind of search generative models are built for, and it's where AI is reshaping the hunt for new drugs most concretely.
But a second thread running through the day's reporting argues the biggest bottleneck isn't the algorithm at all. The models are the visible part; what constrains them sits elsewhere in the pipeline.
Which makes the day's smallest funding story quietly relevant. OmicsBank raised $2.25 million on a simple bet: that the raw material AI drug discovery runs on — patient health data — is about to get a lot more valuable. The company plans to funnel overseas patient data into AI drug discovery. It's a modest round, but it's aimed squarely at the input side of the equation rather than the model side.
AI showed up in the clinic today too. A new study in Nature applies explainable artificial intelligence to prostate surgery, one of the most common cancer operations in men — using AI to look inside the operating room and identify what actually makes the procedure work. "Explainable" is the operative word: the goal is a model whose reasoning surgeons can inspect, not just a prediction.
Three regulatory setbacks, one approval
The FDA had a busy and mostly discouraging day for developers.
The most serious: regulators again halted testing of Regenxbio's experimental gene therapy for Hunter syndrome, a rare inherited disorder, after investigators detected masses on the spines of five patients. This is not the trial's first hold.
Capricor Therapeutics got a softer but still costly delay — the FDA pushed back its decision date on deramiocel, the company's cell therapy for Duchenne muscular dystrophy, buying itself more time to review additional data Capricor submitted. Another pause in a program that has already seen one.
The bright spot belonged to Jazz Pharmaceuticals, which won expanded U.S. approval for Ziihera, its bispecific antibody, in certain gastric cancer patients. The key detail is placement: the clearance covers newly diagnosed patients, moving the drug earlier in the treatment sequence — generally the more valuable position commercially and clinically.
Deals, and one undoing
Bristol Myers Squibb ended its partnership with Cellares, which builds automated manufacturing systems for cell therapies, after what the pharma giant called a "comprehensive evaluation." Cellares is restructuring as a result — a reminder of how much a single anchor partner can matter to a manufacturing-stage company.
On the buy side, Sword Health has proposed acquiring Headspace, the mental health startup, a plan revealed through a Massachusetts state filing rather than an announcement. And Haisco Pharmaceutical, a China-based drugmaker most American readers have never heard of, signed deals bringing two of biotech's best-known names onto its roster — the latest sign of Chinese pharma buying its way into Western credibility.
Digital health consolidates around AI
Two nutrition companies on opposite sides of the world are merging: US-based Berry Street is combining with India's Healthify to build an AI-driven metabolic care business.
Meanwhile, women's health startups are attempting a comeback after a rough stretch. Venture funding to the sector peaked in 2024, then declined last year. Founders and funders are now pitching AI as the route back to growth — a familiar move, and one whose success will depend on the same data question hanging over drug discovery.