The AI architect taking aim at RNAi

Jakob Uszkoreit helped create the transformer architecture — the “T” in ChatGPT — that sparked the generative AI boom. Now he’s trying to do something just as ambitious in drug development, STAT’s Brittany Trang writes. His startup, Inceptive Nucleics, is building biological foundation models that can be applied across a wide range of sequence-based medicines, from RNA interference therapies to mRNA and antisense drugs.

That vision caught the attention of Alnylam, which yesterday announced that it had struck a three-year partnership worth up to $2 billion in potential milestone payments and royalties, along with a $30 million upfront investment. The idea is that AI could perhaps do more than analyze biological data — and instead design the molecules themselves.

liquid biopsy

Why an oncologist says Grail's 'failed' trial still matters

At ASCO, the long-awaited results from a randomized trial of Grail’s Galleri multi-cancer early detection test were presented. And while the test failed to reduce late-stage cancers — the primary endpoint of the study — the results are a little more complicated, and potentially meaningful, Brown University radiation oncologist Sana Raoof opines for STAT.

The study actually revealed several encouraging signals, she writes. The test had a positive predictive value of 52%, which is “quite good” compared to other screening tests that are already approved. Also, it quadrupled the number of screen-detected cancers, reduced the incidence of emergency cancer presentations, and appeared to shift some diagnoses from stage 4 to stage 3 disease.

The problem, she writes, is that Grail chose a difficult endpoint that counted reductions in stage 4 cancers and increases in newly detected stage 3 cancers as offsetting each other, which potentially obscured meaningful clinical benefit. The trial’s short follow-up and methodological challenges make it too early to dismiss multi-cancer early detection tests, which she still sees as a critical future component of cancer screening.

cancer

Does RevMed's Daraxonrasib have greater potential?

Daraxonrasib, the buzzy experimental pancreatic cancer treatment from Revolution Medicines, looks increasingly destined to become the new standard second-line treatment for the disease, STAT’s Adam Feuerstein writes in his weekly Biotech Scorecard. Now, the question is whether the drug can reach newly diagnosed patients.

Speaking at a STAT event during ASCO, former FDA cancer drug regulator Rick Pazdur argued that daraxonrasib’s impressive survival benefit could make traditional randomized trials difficult to conduct, because patients may be unwilling to risk being assigned to chemotherapy alone.

Pazdur floated a novel “reverse accelerated approval” approach: grant earlier approval based on tumor responses in the ongoing Phase 3 trial, then confirm long-term survival benefits using external control data.

“We know [daraxonrasib] is highly active in the first line,” RevMed CEO Mark Goldsmith said. “But what do we need to show to convince the decision-makers? That's something that we just have to learn. We have a terrific relationship with the folks at the FDA. They will have every interest in trying to serve patients as we do. We'll keep working on this.”

artificial intelligence

AI titans push Congress for DNA safeguards

From STAT's Brittany Trang: Sam Altman (CEO of OpenAI), Dario Amodei (CEO of Anthropic), Eric Horvitz (CSO of Microsoft), David Baker (2024 Nobel laureate), and dozens of others have signed an open letter urging Congress to act on the biosecurity threat posed by AI.

"As I have argued, sequence-level screening is necessary but not sufficient — de novo designed proteins may bear little or no homology to known threats. That's why strong records retention is equally important," said Baker in a statement.

The overarching concern is that AI will come up with sequences for biological weapons that current screening software cannot detect, as Horvitz and colleagues demonstrated in a Science paper last year. The signatories want screening processes mandated, and for all businesses that syntheize DNA or manufacture synthesizers to keep logs of what sequences have been ordered in case something slips through screening. A bipartisan bill that would mandate some of these measures was introduced in February.

More around STAT