Verge Genomics relaunched this week as Verge Labs, a “frontier AI lab for human disease biology.” What you don’t see behind the rebranding is the 90% layoff and a drug that didn't work. In December, Verge's AI-designed ALS candidate failed its early trial. The marker for nerve damage went up, not down.

CEO Alice Zhang's comments is fair enough. First attempts rarely produce blockbusters, and traditional drugs fail all the time. Hard to argue with that.

What does stand out is that Verge isn't taking another swing at making a drug. It's becoming a data business, selling targets, patient-matching, and biomarkers to the pharma companies that still carry the risk. Chai Discovery and Noetik have made similar moves toward partnership over going it alone. So there seems to be a running theme here.

The original AI drug discovery pitch was that the models would find better drugs. But what’s actually happening is the fallback leads to just selling data to the incumbents.

It looks to me that the wave of AI drug discovery promises is starting to be tested. Even if these companies that set out to prove AI can discover drugs keep retreating into data services, to me, it’s still a stepping stone towards generating more high-quality (hopefully…) data that will bridge the AI drug to market gap.

For more details: Full Article

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

Eikon Therapeutics has fully enrolled its Phase 2 trial of EIK1001 in first-line stage 4 lung cancer and will read out interim data at ASCO starting May 29. The TeLuRide-005 study pairs EIK1001, an immune-activating TLR7/8 agonist, with pembrolizumab and chemotherapy, and the update covers a full interim look at the non-squamous cohort plus a partial look at the squamous cohort. [Link]

Biohub just launched an open-source "world model" of protein biology to accelerate drug discovery. The Chan-Zuckerberg philanthropic venture says the system learns from the protein sequences produced by evolution, then uses that knowledge to design new proteins. Researchers have already used the models to create protein binders for cancer and immune targets that reactivated immune cells in lab tests. [Link]

The FDA's 2026 guidance agenda includes new rules for using AI and machine learning in pharmaceutical manufacturing. CDER's agenda lists 81 guidelines scheduled for release this year, with a new addition specifically covering AI/ML quality considerations in pharma manufacturing. The agenda also adds guidance on digital health technologies in clinical investigations and distributed manufacturing. [Link]

A Penn study found that online patient-facing information about AI in cancer care is sparse, low quality, and often silent on the risks. The cross-sectional analysis, being presented at ASCO, reviewed available webpages and videos and concluded oncology groups have an opening to build clearer, higher-quality resources as patients increasingly turn to AI to understand their diagnoses. [Link]

A woman with epilepsy finds out she's pregnant. She's taken the same drug for years, and now she has to choose whether to stay on it and worry about the baby, or stop and risk a seizure. She asks her doctor what the medication does during pregnancy, and the honest answer is some version of “we don't really know.”

That conversation happens millions of times a year. Around 79% of women take at least one medication while pregnant, and almost none of those drugs were ever tested in anyone who was.

After thalidomide caused thousands of birth defects in the late 1950s, the FDA moved in 1977 to keep any woman “capable of becoming pregnant” out of early clinical trials.

The instinct was protective, but the result is a major knowledge loss in hindsight, because thalidomide is exactly the kind of drug a trial would have caught.

One analysis estimated that a 200-person study would have harmed about 33 children while preventing roughly 8,000 birth defects. The disaster that pushed pregnant women out of research is the disaster that studying them would have prevented, and the reflex outlived its reason. A 2025 analysis found that fewer than 1% of US drug trials still enroll pregnant participants.

Pregnant women keep taking medications regardless, and every one of those exposures, along with whatever happened next, ends up in insurance claims and medical records.

Pregnancy safety information, where any exists, comes mostly from this post-approval real-world rather than from trials. In 2025, the FDA issued draft guidance on bringing pregnant and breastfeeding women into clinical trials, a sign that regulators now want precisely this kind of evidence.

Almut Winterstein's BOOST-HP project reads those records at scale, hunting for links between drugs taken in pregnancy and the outcomes that followed, then sorting statistical noise from the signals worth chasing.

Cristina Longo's BIONIC study points machine learning at a narrower question: which pregnant women on biologic drugs face the highest risk of harm.

In most machine learning, a wrong prediction is a number on a leaderboard. Here, it reaches a real woman deciding whether to keep taking a drug while she's pregnant. That is why neither group reaches for the most powerful black-box model available.

Winterstein keeps her models fully traceable because she's watched AI systems break basic rules of epidemiology, like implying a drug caused something that happened before the woman ever took it.

The records these models lean on are messy and full of gaps. And these are early projects, described in a magazine survey rather than proven out in a trial, often missing the demographic detail needed to keep the predictions fair.

Pull real safety signals out of records that already exist, and pregnant women could enter trials earlier and more safely, which would build better evidence still.

For fifty years, the field treated knowing nothing about pregnant women as the careful choice. The consequences of that caution never disappeared.

Sources: [Research Article]

Despite accounting for roughly 10% of total brain volume, it is densely packed with tiny neurons (specifically granule cells) and contains over 50% (and by some estimates, up to 80%) of the brain's total neurons.