WISER × Moderna · Quantum+AI 2026
We ask a quantum optimizer to predict how a short strand of RNA folds, then hold its answer up against the classical gold standard. Sometimes it nails it. Sometimes it doesn't — and that gap is the interesting part.
01 · The problem
A strand of RNA doesn't stay a straight line. It folds back and pairs with itself, and the shape it settles into decides how it behaves.
For an mRNA medicine, that folded shape influences how stable the molecule is, how efficiently a cell can read it to make protein, and how easy it is to manufacture. Predict the shape well, and you can design better medicines. The catch: the number of possible ways a strand can fold explodes as it gets longer, far too many to check one by one.
Tools like ViennaRNA use decades of thermodynamics to find the single lowest-energy fold — the “minimum free energy” shape. It's fast and trusted. We treat its answer as ground truth.
Can a quantum optimizer rediscover that same shape by searching the energy landscape a different way? Not to beat the classical tool today — but to learn how the approach scales, and where it breaks.
02 · How to read this
Lay the RNA letters out in a line. Each time two letters pair up, we draw an arc connecting them. A folded structure is just a collection of these arcs. Two predictions agree when their arcs match.
In the explorer below, we stack our quantum method's answer above the line and ViennaRNA's answer below it. When the two mirror each other perfectly, the prediction is exact. When the top has arcs the bottom doesn't, the quantum method predicted a pairing that shouldn't be there.
03 · The results
Pick a strand. Green arcs are pairings both methods agree on; red arcs are ones only the quantum method predicted; grey dashed arcs are ones only ViennaRNA found.
04 · The honest part
On two of our test strands, the quantum method got the wrong answer. So we checked something: could a brute-force search that tries every possibility do better?
It couldn't. Brute force landed on the same wrong answer. That tells us something precise: the mistake isn't the quantum optimizer being weak — it's that our simplified energy model rewards a pairing that real thermodynamics wouldn't. The optimizer solved the problem we gave it faithfully; the problem itself was slightly off. That's a formulation gap, not a solver failure — and knowing the difference is the whole point of a benchmark.
05 · Real hardware is noisy
Real quantum machines make errors. We simulated that — running the smallest sequence under increasing amounts of sampling noise and hardware-inspired gate noise — to see whether the correct fold falls apart.