AI is finally breaking open cancer's most stubborn targets

For decades, cancer researchers have stared at a handful of deadly proteins and thrown up their hands.

Proteins like RAS, p53, and MYC drive the vast majority of human tumors. But because their molecular surfaces are completely smooth, traditional small-molecule drugs simply slide off, earning them the frustrating label of "undruggable."

Now, that stubborn wall is finally cracking.

Reports highlighted breakthrough clinical data for daraxonrasib, a new pill developed by Redwood City-based Revolution Medicines. In a global trial of 500 advanced pancreatic cancer patients who had failed standard therapies, the drug nearly doubled median survival time from 6.6 months on chemotherapy to 13.2 months.

Patients on the pill also reported significantly less pain and better quality of life, with rash being the most common side effect.

Hitching a ride on the switch

RAS genes were discovered way back in 1982. When functioning properly, RAS acts as a molecular switch for cell growth, turning on and quickly shutting off. Mutant RAS gets jammed in the "on" position, fueling uncontrolled tumor growth.

While the first RAS-targeting drug won U.S. approval in 2021, it only blocked one single mutation in one of three family members. Daraxonrasib does something much bolder: it shuts down all three RAS family proteins.

Instead of trying to find a grip on the slippery protein itself, the drug binds to a molecular partner that assists RAS with folding. By hitching a ride on this partner, it wedges itself into the active RAS complex and shuts it down.

AI hunts down p53 and MYC

Similar breakthroughs are accelerating across other historically untouchable targets, largely thanks to machine learning.

Take p53, the "guardian of the genome" regulating over 300 genes. A team at Baylor College of Medicine deployed AI to screen roughly 10 million chemical compounds, hunting for molecules that prompt mutated p53 cells to self-destruct while leaving healthy tissue intact. The AI surfaced 83 candidates, including one dubbed H3 that dramatically curbed tumor growth in mice.

Meanwhile, researchers tackling MYC—an uncontrolled growth regulator implicated in roughly 70 percent of cancers—are deploying AI models to map out molecular binding pockets and find compounds that hamstring tumor DNA repair.

Why it matters

In our view, daraxonrasib is proof that medicinal chemistry workarounds actually work in the clinic, while AI is ready to scale that success across oncology.

For advanced pancreatic cancer patients facing dismal prognoses, doubling survival to over 13 months with manageable side effects is genuinely life-changing. Just as critically, a daily oral pill is vastly cheaper, easier to distribute, and more accessible than personalized CAR T cell infusions, which still struggle against solid tumors.

For the biotech industry, combining structural biology tricks with AI-driven screening of millions of compounds flips the drug discovery pipeline on its head. The traditional concept of "undruggable" proteins is expiring.

That said, it is not an outright cure. Cancer is notoriously adaptable, and secondary resistance mechanisms will likely emerge over time. But by proving that pan-RAS inhibition works and showing that AI can rapidly unearth hidden binding strategies, medicine has taken its most confident step yet into cancer's hardest territory.