Daraxonrasib, a drug that targets the entire RAS cancer family, added roughly half a year of life in advanced pancreatic cancer and signals a new playbook for 'undruggable' targets.
When daraxonrasib was added to chemotherapy for advanced pancreatic cancer, the trial reported that patients lived a median of 13.2 months, roughly double the 6.6 months seen with chemotherapy alone, and the drug caused fewer severe side effects, according to the New England Journal of Medicine. For a disease with one of the worst prognoses in oncology, half a year of additional life is not a footnote. It is the kind of checkpoint that resets how oncologists and patients talk about options.
The result is the first clinical payoff from a 40-year effort to drug the RAS family of proteins, which sit at the top of the signaling cascade that tells cells when to grow. KRAS, NRAS, and HRAS mutations drive roughly a third of all human cancers, including most pancreatic tumors, and they have been called "undruggable" since the RAS oncogene was first identified in 1982. The reason was shape. RAS proteins are smooth, with no obvious pocket for a small molecule to grip, and the one crevice that did work, the G12C mutation, took until 2021 to yield the first US approval: sotorasib, marketed as Lumakras by Amgen, for a narrow slice of lung cancer patients.
Daraxonrasib, developed by Revolution Medicines in Redwood City, California, takes a different approach. Instead of trying to grab the smooth surface of RAS directly, it forms what chemists call a "tri-complex": the drug binds to a RAS partner protein and pulls RAS into a three-molecule structure that does have a usable binding pocket. The chemistry is laid out in a Journal of Medicinal Chemistry profile of the molecule. The practical consequence is that daraxonrasib is a pan-RAS(ON) inhibitor, meaning it can hit HRAS, NRAS, and KRAS alike, including the KRAS G12D and G12V variants that dominate pancreatic cancer, which sotorasib and adagrasib were never designed to touch.
That broader reach is what made the pancreatic trial possible. Pancreatic cancer is the test case because the disease is overwhelmingly KRAS-driven, with no major subtype distinctions like the G12C-only patients who could get sotorasib. In the trial, the most common side effect of daraxonrasib was rash, not the bone-marrow suppression and neuropathy that chemotherapy causes. The patients were sicker by definition, all with advanced disease, and the survival gain is still measured in months, not years. Resistance is a near-certainty on the timeline oncologists care about. Daraxonrasib is not a cure. It is, however, a working drug for a target that spent four decades rejecting every attempt to drug it.
The same playbook, target a smooth, featureless surface by recruiting the protein's own partners, is now being turned on the next undruggable candidate: p53, the tumor-suppressor gene called "the guardian of the genome" and first described in 1979. Roughly half of all cancers disable p53, but the protein is small and lacks the kind of active site a small molecule can grip. The new strategy is to bind the mutant protein directly and pull it back into a working shape. An early trial of a p53 restabilizer, published in NEJM in early 2026, reported tumor shrinkage in roughly 20 percent of patients within 21 days, across ovarian, breast, and other solid tumors, with the trial registered as NCT04585750. The numbers are small and the patient population is heterogeneous, so the result reads as a signal, not a verdict.
Further out, researchers at Baylor College of Medicine used AI to screen roughly 10 million compounds for synthetic-lethal killing of mutant p53 cells. The screen returned 83 chemically distinct hits, and the lead compound, called H3, suppressed tumor growth in mouse models. That is preclinical, mouse-only data, and the long path from a mouse xenograft to a Phase 1 trial is where most of these candidates disappear.
MYC is the next frontier after that. Abnormal MYC activity is implicated in roughly 70 percent of cancers, and MYC has resisted every direct-inhibitor attempt for similar reasons as RAS. No human trial data on a MYC-targeted agent has yet matched the daraxonrasib story.
AI's role in the current wave, as the Singularity Hub synthesis of the field frames it, is as an accelerant rather than an inventor. Structure-prediction systems such as AlphaFold and a new generation of generative protein-design models can predict how a candidate will fold into a target, screen millions of compounds against a binding site, and propose new structures. They did not invent daraxonrasib or the tri-complex idea. The clinical effect came from chemistry, medicinal optimization, and patients who agreed to enroll in a trial for a disease with a historically grim prognosis. The 13.2-month number is the result. The rest is the question of how many more targets can be cracked the same way.