The algorithm was never the hard part
The loudest story pharma tells about itself right now is an AI story: models that propose new molecules, predict how they behave, and compress years of bench work into months. Today's reporting keeps poking at the same soft spot in that pitch — the models are ahead of the material they run on.
Start at the beginning of the pipeline. "Hit identification" — picking the handful of molecules worth testing out of an almost unimaginable number of possibilities — is a needle-in-a-haystack problem, and generative AI is genuinely reshaping how that search gets done. But a second piece out today makes the counterpoint plainly: the biggest bottleneck isn't the algorithm. It's everything upstream of it.
Two other items today are effectively that argument in practice. OmicsBank raised $2.25 million on the bet that the raw material of AI drug discovery — patient health data, sourced overseas — is about to become far more valuable, per Endpoints News. And the nonprofit Target ALS published a piece, surfaced via Google News, on building the data foundation for AI-driven ALS drug discovery. Different scales, same thesis: before AI can cure anything, someone has to do the unglamorous work of assembling the data. Read together, today's four AI stories form a single arc — enormous promise at the top of the funnel, and a scramble to supply it.
Approvals and a very generous reading of a miss
The day's cleanest piece of hard news: the FDA cleared Jazz Pharmaceuticals' bispecific antibody Ziihera for use in certain newly diagnosed gastric cancer patients, an expanded U.S. approval reported by Endpoints News. Moving a drug earlier into the treatment sequence — to newly diagnosed patients — is the kind of label change that meaningfully widens a commercial opportunity.
Less clean: Bausch + Lomb's experimental dry eye treatment missed its target in a mid-stage study, and the company is advancing to a late-stage trial anyway. The CEO told Endpoints exclusively that they view the result as a win. Companies are entitled to see signal where a primary endpoint didn't, but "failed Phase 2, proceeding to Phase 3" is a phrase that deserves to be read twice.
Partnerships made and unmade
Bristol Myers Squibb has ended its partnership with Cellares, which builds automated manufacturing systems for cell therapies. Endpoints News reports the pharma giant made the call after a comprehensive evaluation, and the fallout is forcing Cellares into a restructuring — a reminder of how much oxygen a single big-pharma partner supplies to a manufacturing-tech startup.
Cutting the other way: Haisco Pharmaceutical, a China-based drugmaker most American readers have never heard of, has signed deals bringing two of biotech's best-known names into its orbit, according to Endpoints News. The pattern of Western biotech talent and capital flowing toward Chinese drugmakers keeps repeating, and Haisco just became the latest data point.
Where the money went, and didn't
Venture investment in women's health startups peaked in 2024 and then declined last year, BioPharma Dive reports, leaving founders and funders looking for a way back. The proposed comeback route is, predictably, AI — which is either a real efficiency unlock for an underfunded field or the same story every underfunded field is currently telling. Watch which it turns out to be.
The frontier file
Three items that sit further out. Researchers have used generative AI to design viruses that kill drug-resistant E. coli, per AZoRobotics via Google News — AI pointed at antibiotic resistance, one of medicine's genuinely hard problems. Seoul St. Mary's Hospital has turned to quantum computing for heart disease research, per Chosunbiz, a rare case of a hospital rather than a lab making that bet. And a new paper in the peer-reviewed journal Cureus examines AI entering psychiatric training — the field weighing what automation costs in medicine's most human-centered specialty.
The through-line for the day: the tools are arriving faster than the foundations, the data, and the norms they need.