Generate:Biomedicines just pulled off the biggest biotech IPO of 2026, raising $400 million on the promise of AI-designed protein therapeutics. The stock promptly dropped 21% soon after...

Welcome to Wall Street, I suppose.

Here's the thing, Generate's lead program, an anti-TSLP antibody heading into Phase III for severe asthma, isn't a de novo AI creation. It's an optimized variant of a known protein against a validated target.

The CEO himself compared it to being built on "the equivalent of the original ChatGPT." Meanwhile, the truly novel stuff (antibodies designed from scratch by AI) hasn't reached a single patient yet.

That's the tension underneath this whole IPO wave. Biotech offerings are up 53% year-over-year. Investors are pricing in a future where AI reinvents drug discovery. But the clinical pipelines backing these valuations are still largely conventional molecules with an AI assist. (See our special report on what makes an AI drug).

So what exactly is the market buying? The drugs we can see, or the platform we're told is coming?

For more details: Full Article

Compugen: The company reported quarterly results that beat expectations this week, hence the nice little bump in their stocks.

Algorae Pharmaceuticals: The company applied to list ~5.06 million new shares on the ASX, which means dilution risk for current holders.

For the rest, no real news, but still some major changes…

Select the right answer! (See explanation below and source)

Eikon Therapeutics is now recruiting for a Phase 1/2 trial of EIK1005, its single-molecule-tracking-derived oncology candidate, in combination with Merck's Keytruda. The study targets advanced solid tumors including MSI-high and mismatch repair deficient cancers across gastric, colorectal, and endometrial indications. Eikon's platform uses super-resolution microscopy to track individual protein movements in living cells, a unique approach among AI-era drug discovery companies. [Link]

A STAT op-ed argues that AI is generating promising antibiotic candidates at unprecedented scale, but the broken antibiotic market means most never get synthesized or tested. The authors, a UPenn professor and the CEO of the $1 billion AMR Action Fund, point to a Nature study where AI predicted hundreds of promising molecules but only two dozen could be tested because no partners would fund the work. They call for subscription-style reimbursement and market-entry rewards to fix the downstream economics. [Link]

Tempus expanded its collaboration with Merck in a multi-year deal giving Merck access to Tempus' massive multimodal dataset and GPU-powered Lens Platform for biomarker discovery. The partnership focuses on precision oncology, including identifying resistance mechanisms and informing drug combinations for Merck's early pipeline. Tempus continues to position itself as the data and compute infrastructure layer for pharma's AI efforts. [Link]

Regeneron joined the Alliance for Genomic Discovery as its tenth pharma member, expanding the dataset to 312,000 whole genomes paired with deep clinical data. The alliance, run by Illumina and Nashville Biosciences, also launched a new 50,000-genome proteomics layer using Illumina's Protein Prep technology, with GSK among the first participants. Members now include AbbVie, Alnylam, Amgen, AstraZeneca, Bayer, BMS, GSK, Merck, Novo Nordisk, and Regeneron. [Link]

Do you remember Google Flu Trend? The Google project that tried to predict flu outbreaks by tracking what people searched for, then famously overestimated flu prevalence by 140% during the 2012-2013 season before being shut down.

Turns out the problem wasn't using digital signals for epidemic surveillance. It was using the wrong digital signals. A team from Peking Union Medical College and the Chinese CDC just published a paper in npj Digital Medicine that makes a surprisingly elegant argument: if you want to know who has the flu, don't watch what people Google. Watch what they buy.

The researchers analyzed 21.08 million digital prescription transactions for flu antivirals (oseltamivir and baloxavir marboxil) from Meituan Health, China's dominant online pharmacy platform, across all 31 provinces from 2022 to 2024.

That's every time someone saw a doctor online, got a prescription, and had Tamiflu delivered to their door, often within an hour. And here's what separates this from every other "we correlated digital data with disease" paper: they didn't just show correlation. They showed causation.

Using a technique called convergent cross-mapping (a method for detecting causal relationships in complex systems), they found that prescription data and lab-confirmed flu positivity are bidirectionally coupled.

Prescriptions predicted flu activity in 28 of 31 provinces, and flu activity predicted prescriptions in 22 of 31. That two-way relationship is the whole ballgame. It means the prescription signal is genuinely tracking the epidemic, not just riding the same seasonal wave.

Online search data, by comparison, showed strong forward prediction (30/31 provinces) but almost no reverse causality (2/31). Search reflects panic as much as it reflects illness.

Prescriptions require a doctor, a diagnosis, and a credit card.

The team then built a deep learning model combining graph neural networks, Mamba state-space models, and LSTMs to forecast daily prescription rates. It achieved reliable 96-day predictions in 29 of 31 provinces, covering an entire flu season.

That's not a two-week nowcast. That's a three-month advance warning for vaccine campaigns, hospital staffing, and antiviral stockpiling.

Although there are some things to consider. Meituan Health's user base skews young and urban, which means the elderly and rural populations most vulnerable to severe flu are underrepresented.

The data only captures prescription antivirals, missing the majority of mild cases where people just ride it out with over-the-counter meds. And China's digital pharmacy ecosystem, where a single platform controls roughly 70% of the online pharmacy market share and delivers prescriptions within the hour, doesn't exist in most countries.

Whether this approach transfers to the fragmented pharmacy landscapes of the U.S. or Europe is an open question.

Still, the underlying logic is hard to argue with. The same principle of using pharmacy transaction data for flu surveillance has shown up in studies from Japan, the U.S., and Canada using traditional brick-and-mortar sales.

The digital version just makes it faster, with 24-hour data availability instead of weekly or monthly reports. And as wastewater surveillance gains traction as another real-time monitoring tool, the case for building multi-source epidemic detection systems gets stronger.

What sticks with me is how this reframes a decade-old failure. Google Flu Trends collapsed because searching for "flu symptoms" is free, reflexive, and driven by media cycles.

Buying prescription antivirals requires clearing a series of real barriers: you feel sick enough to act, a doctor agrees, and you pay. Every one of those steps filters out noise.

Sources: [Research Article]

Studies have reported a small increase in heart attacks (myocardial infarctions) during the first days after the spring clock change. The likely contributors include sleep deprivation, circadian rhythm disruption, and increased sympathetic nervous system activity. [Source]