The FDA’s August approval of the first mRNA flu vaccine for adults 50 and up marked a milestone: RNA therapeutics have moved from pandemic emergency into routine care. But Stanford biochemist Rhiju Das says the real shift is just beginning. In a wide-ranging interview, he outlines why RNA could become the dominant class of medicine, how artificial intelligence solved a folding problem that stumped humans for decades, and what it will take to turn AI-designed molecules into approved drugs. [1] [4]
Why RNA could displace proteins and small molecules
Das has watched the RNA renaissance for 25 years. He argues RNA therapeutics are “more sophisticated and adaptable” than protein-, cell-, or small-molecule-based medicines because a single molecule can carry genetic instructions, regulate its own activity, and respond to cellular context. That versatility means RNA may eventually compete with or displace entire categories of existing drugs. The mRNA flu vaccine and the COVID-19 shots before it proved the platform works at scale; the next step is expanding what RNA can do inside the body.
The folding bottleneck and how AI broke it
Stability has been the practical headache. Early RNA vaccines required deep-freeze supply chains because unfolded RNA degrades quickly. Das hypothesized that tightly folded RNA would be shelf-stable at warmer temperatures — a property natural folded RNAs already exhibit. His lab crowdsourced the design problem through Eterna, a video-game platform where 15 years of players have produced nearly 40 peer-reviewed publications. Humans got good at designing complex 3D RNA shapes, but then AI teams entered the challenge. After a few rounds, aggregated AI designs matched or exceeded human performance — the first time a computational method beat experienced humans at RNA design. All four AI-generated designs yielded clear, atomically ordered structures under cryo-electron microscopy, including folds never seen in nature.
Programmable medicines that sense and compute
Those novel folds are more than structural curiosities. Das says the sophistication means future RNA therapeutics could “sense what kind of cell they’re in, compute molecular signals, and even control when a treatment turns on or off.” In other words, RNA becomes a molecular logic circuit: a drug that activates only in diseased tissue, adjusts dosage based on biomarker levels, or self-limits to reduce side effects. That programmability is what separates RNA from static small molecules or proteins.
The five-year roadmap and how to join
Despite the breakthrough, Das estimates AI-designed RNA drugs are “at least another five years away.” The field still needs to solve several foundational design problems — challenges his team and the broader community are tackling now on Eterna. Those solutions will become the seminal proofs of concept that seed a new RNA-design ecosystem. Crucially, participation doesn’t require a PhD. Eterna’s game interface lets anyone contribute designs that are experimentally validated in the lab. Das calls it a rare ground-floor opportunity in a technology poised for massive impact.
Parallel track: AI for better flu vaccines
While RNA therapeutics mature, AI is already reshaping the seasonal flu shot itself. MIT’s VaxSeer system combines evolutionary modeling with antigenicity predictions to forecast which strains will dominate, and retrospective testing shows it would have outperformed WHO strain choices over the past decade. Meanwhile, frameworks like MAIVeSS have cut strain-selection time from months to days, and an AI-designed vaccine from Flinders University entered U.S. trials as early as 2019. These efforts don’t replace RNA platforms; they improve the antigens those platforms deliver.
The convergence is clear: AI is learning to write the genetic code for medicines that fold, function, and decide. The flu vaccine was the proof of concept. The programmable RNA pharmacy comes next.