Chalmers, Lund and Jülich Designed Lattice-Protein Sequences on IBM's Torino Quantum Processor: What Physical Review Applied Published on 28 September 2026
Quentir Medicine Monitor
Evidence-based insights for quantum medicine. Published by Quentir Systems LLC · October 1, 2026.
Most drug-discovery stories about quantum computers ask a folding question: given a chain of amino acids, which shape will it settle into? Protein designers work the other way round. They start from a shape they want, a pocket that grips a drug or a surface that binds a virus, and search for a sequence of building blocks that will fold into it. A team from Chalmers University of Technology in Gothenburg, Lund University and the Jülich Supercomputing Center has now measured how far one of today's quantum processors gets on the first half of that design problem.
Their paper, "Designing lattice proteins with variational quantum algorithms", appeared in Physical Review Applied (volume 26, article 034065) on 28 September 2026, a year after a first preprint on arXiv in August 2025. Hanna Linn, Lucas Knuthson, Anders Irbäck, Sandipan Mohanty, Laura García-Álvarez and Göran Johansson compare two families of variational quantum algorithms on a task they call protein sequence optimization, and they run the more robust family on IBM's Torino quantum processor, a device built on IBM's Heron r1 chip. With circuit settings carried over from noiseless computer simulations, the hardware found the correct sequence in more than 10 percent of its attempts for 17 of the 22 test problems, including the three largest, chains of 27 to 29 units.
The test proteins are simplified models drawn on a flat grid, and the authors say so plainly. The paper is still a careful read for anyone who funds or buys computational chemistry. In one controlled setting, with every right answer known in advance, it shows where current quantum hardware helps on a design task, where it breaks down, and which engineering choices made the difference.
The test bed is the HP model, one of the oldest simplifications in protein physics. Each protein is a chain of beads on a two-dimensional square grid, and every bead is one of only two kinds. H beads are hydrophobic, the water-avoiding kind of amino acid that real proteins tend to bury in their core. P beads are polar. The energy of a folded chain drops by one unit for every pair of H beads that touch on the grid without being direct neighbors along the chain, so the model rewards a compact hydrophobic core. Real proteins have twenty kinds of amino acid, side chains and three dimensions, so the HP model is a laboratory for methods. Its great advantage is that for chains of up to 30 beads, exhaustive classical enumeration has already listed the exact answers, which lets researchers grade a new algorithm without guessing.
What Linn, Knuthson, Irbäck, Mohanty, García-Álvarez and Johansson asked IBM Torino to compute
Protein design is usually split into two steps. The first searches for sequences that have the lowest possible energy in the target shape. The second checks whether those sequences really fold into that shape and not into some competing one. The Gothenburg, Lund and Jülich team studied only the first step. Each problem also fixes how many H beads the sequence must contain. Without that rule the answer would be trivial, because a chain made only of H beads collects the most H-H contacts. The paper enforces the count with a penalty term in the energy function, and some of its QAOA variants keep the count fixed inside the circuit itself. The team picked one target shape and one H-bead count for each of 22 selected chain lengths between 4 and 29 beads, so that each problem has a single best sequence that is already known to fold into the target. According to the exhaustive enumerations the paper cites, only about 2 percent of all HP sequences up to 30 beads have such a unique lowest-energy shape, so the selected targets are the most designable shapes for their length.
The quantum encoding is economical. Each qubit decides whether one position in the chain is H or P, so a chain of N beads needs only N qubits. Folding needs considerably more, because the computer must also represent where every bead sits on the grid. That difference explains why design is a sensible early target for small, noisy machines. The Monitor saw the folding side of the same field in IonQ and Kipu Quantum's 64-qubit protein-folding paper in September, where chains of 14 to 16 residues already needed 46 to 61 qubits.
Quantum pillar: computing. Technology readiness: TRL 3 of 9. The method ran on a real IBM quantum processor as an experimental proof of concept, on simplified two-letter model proteins whose correct answers were already known, many steps away from designing an actual drug, enzyme or antibody.
Why the QAOA circuits failed under noise and the hardware-efficient circuits held up
The first family the team tested is the quantum approximate optimization algorithm, QAOA, in five variants. QAOA builds the problem itself into the quantum circuit, which is elegant on paper. In noiseless simulations of chains up to 16 beads, the strongest variant, which combines a fully connected mixing step with a carefully prepared starting state, found the right sequence with a probability of at least 0.95 every time. Once the authors added a noise model derived from the Torino device, the success rate of all five variants fell sharply. The reason is circuit depth: that strongest variant needed more than 2,000 layers of operations for a 16-bead chain, and every layer gives noise another chance to corrupt the answer. The authors conclude that QAOA in its current form does not suit today's hardware for this problem.
The second family is the hardware-efficient ansatz. Its circuit ignores the structure of the protein problem and uses only the gates and qubit connections the device handles well, in one or two short layers. In perfect simulations it did worse than the strongest QAOA variant. Under realistic noise it did better, and the single-layer version proved the most noise-tolerant of everything tested. One further trick mattered for the longer chains. Starting each problem from random settings gave poor results, so the team passed the tuned settings of a smaller problem on to the next larger one, a scheme the paper calls parameter donation.
What 20 experiments of 10,000 shots each showed on IBM Torino for chains of 4 to 29 beads
For the hardware runs, the team fixed the circuit settings taken from their most successful simulation runs and executed each problem 20 times, with 10,000 measurements per experiment. With the two-layer circuit and settings from the noiseless simulations, success rates exceeded 10 percent for 17 of the 22 instances, including the three largest chains of 27, 28 and 29 beads. The problems that stayed below that threshold were the same ones on which the simulations had already done poorly, and the one-layer circuit fell short more often, on 7 of the 22. The hardware success rates came out somewhat lower than the simulated ones with identical settings. The authors read that gap as evidence that their noise model misses error sources such as drift over time and correlated errors across several qubits. They also note that difficulty does not rise smoothly with chain length: the 17-bead problem proved harder than its size alone would suggest, because the shape of the target and the number of H beads matter too.
The group has context for these numbers. In a 2024 study in Physical Review Research, Irbäck, Knuthson, Mohanty and Carsten Peterson attacked the same design problem on a D-Wave quantum annealer. The purely quantum solver saw its success rate fall roughly exponentially with chain size, to about 1 to 10 percent at 20 beads. D-Wave's hybrid solver, which splits the problem into smaller pieces for the quantum chip, handled chains up to 64 beads and compared favorably with classical Monte Carlo methods. The new paper suggests that gate-based machines could gain from the same kind of problem splitting and from closer pairing with classical pre- and post-processing.
What the September 2026 result means for drug-discovery teams and hospital research buyers
The paper makes no claim of quantum advantage, and none is available here. Every answer it found was known beforehand from classical enumeration, and the study does not compare running time or cost against a classical solver. Its value lies in method. It shows, with numbers, that circuits shaped to the hardware beat circuits shaped to the problem on today's noisy machines, and that reusing tuned settings across related problems keeps longer chains within reach. Those are lessons a pharmaceutical computational team can carry into its own pilots before it signs a cloud contract for quantum time.
For a hospital or academic medical center, there is nothing to procure on the strength of this work. The sensible watch-points are concrete: a run of this method on three-dimensional lattices, on more than two kinds of amino acid, on chains long enough to resemble a real peptide, and in a head-to-head comparison with the classical design software that protein engineers already use. The research was supported by the Knut and Alice Wallenberg Foundation through the Wallenberg Center for Quantum Technology, and the authors have made their analysis code public, which makes those next comparisons easier for any group that wants to try.
Sources
Primary source: Hanna Linn, Lucas Knuthson, Anders Irbäck, Sandipan Mohanty, Laura García-Álvarez and Göran Johansson (Chalmers University of Technology, Lund University and Jülich Supercomputing Center), "Designing lattice proteins with variational quantum algorithms," Physical Review Applied 26, 034065, published 28 September 2026, with its arXiv preprint 2508.02369. Also drawn on: the 2024 quantum-annealing study by Irbäck, Knuthson, Mohanty and Peterson in Physical Review Research; the readiness assessment is this Monitor's own.