This is a special post for quick takes by HedonicEscalator. Only they can create top-level comments. Comments here also appear on the Quick Takes page and All Posts page.
Two exciting recent innovations on pancreatic cancer!
First, the personalized mRNA vaccine autogene cevumeran, developed by BioNTech and Genentech, just reported 6-year follow-up results from their Phase 1 clinical trial. 16 patients were treated, 8 were responders (showed signs of immune reaction to vaccine), 8 were non-responders.
7/8 responders (87.5%) survived 6 years after surgery, 2/8 nonresponders survived (25%).
AACR meeting notes https://www.aacr.org/blog/2026/04/20/live-updates-from-the-aacr-annual-meeting-2026-monday-april-20/
The most important result in this small trial is that vaccine response is strongly correlated with better outcomes. But for context, the trial was restricted to patients with operable pancreatic cancer. Patients diagnosed with stage 1 or 2 pancreatic cancer have a 5-year survival rate of 12%. Patients who get their pancreatic cancer surgically removed have a 5-year post-surgery survival rate of 20%. This makes the overall 6-year post-surgery survival rate of 56% among the 16 trial patients pretty impressive. Keep in mind that the trial patients may have been healthier than average for other reasons, and small n is small n, so we shouldn't be too hasty until we see Phase 2 and 3 data.
Source on survival rates https://www.pancreaticcancer.org.uk/information/just-diagnosed-with-pancreatic-cancer/if-you-can-have-surgery-to-remove-the-cancer-early-pancreatic-cancer/prognosis-if-you-can-have-surgery/
Second, a small molecule drug I almost missed in the mRNA hype but arguably even cooler, the tri-complex ras inhibitor (!!!) daraxonrasib, developed by Revolution Medicines. Ras is a protein involved in many cancers, with PDAC (the most common pancreatic cancer) being especially dependent on ras, but it has historically been considered impossible to target due to its chemical properties. Daraxonrasib is, as far as I'm aware, the first drug to target generic forms of ras. It does so with an exotic "tri-complex" strategy involving gluing a different protein, cyclophilin A, to ras in order to disable it. Crazy stuff!
Phase 3 results found that daraxonrasib doubled survival time among patients with metastatic pancreatic cancer, 6.7 months to 13.2 months (p < 0.0001). Side effects are kind of nasty, but "well tolerated, with a manageable safety profile" by advanced cancer standards.
Revolution Medicines announcement https://ir.revmed.com/news-releases/news-release-details/daraxonrasib-demonstrates-unprecedented-overall-survival-benefit
Derek Lowe coverage https://www.science.org/content/blog-post/progress-against-pancreatic-cancer-part-one
Lecture by executive/scientist at Revolution Medicines https://www.youtube.com/watch?v=bU3IwuDJx24
Many of the incredible advancements we've made in oncology has been won slowly through another pathway we can target, another drug buying a few more months, another targeted technique to mitigate side effects. Progress is made one step at a time, and these are big steps. I'm excited for the future of biotech.
The AI angle, since many people are hyping the "AI for pancreatic cancer" line:
Yes, deep learning is used in the development of the mRNA vaccine. NetMHCpan, a small neural network, is used to help select immunogenic neoantigens (choose the most promising mutant proteins to target) for autogene cevumeran.
Autogene cevumeran paper https://www.nature.com/articles/s41591-024-03334-7
Most of the computational pipeline detailed in the paper consists of traditional tools, in line with what I've written about AI and personalized mRNA vaccines before. Deep learning is extremely useful for some things, but it should be understood as a specialized tool, not a silver bullet, for now.
AI & mRNA blog https://hedonicescalator.substack.com/p/did-paul-conyngham-really-use-ai
I'm not sure if deep learning was used in the development of daraxonrasib. A brief glance at the paper and previous work shows plenty of references to traditional computational tools, but nothing that stands out to me as modern DL.
Daraxonrasib paper https://pubs.acs.org/doi/full/10.1021/acs.jmedchem.4c02314
Previous work https://pmc.ncbi.nlm.nih.gov/articles/PMC10474815/
The company behind daraxonrasib, Revolution Medicines, is quite enthusiastic about ML. They recently made a deal with AI drug discovery platform Iambic Therapeutics. I don't doubt Iambic's tools will soon prove useful, but it's fair to say "deep learning for drug discovery" is still in the early stages of development.
Iambic announcement http://iambic.ai/post/revolution-medicines-collaboration