Drug designers working on protein-level chemistry have long been blocked by a hard computational wall: classical supercomputers cannot fully model the electronic structure of large biomolecules. Cleveland Clinic, RIKEN, and IBM have now pushed past that wall, running quantum-classical simulations on a protein-ligand complex containing 12,635 atoms, the largest such system ever modeled with quantum computing methods. The result, achieved through a hybrid workflow that paired IBM quantum processors with some of the world’s fastest classical supercomputers, signals a concrete shift in how pharmaceutical research teams may approach molecular modeling at biologically relevant scales.
A 12,635-atom simulation and the hardware behind it
The research team performed electronic structure calculations on two protein-ligand complexes, one containing 11,608 atoms and the other containing 12,635 atoms. Quantum sampling ran on two 156-qubit IBM processors designated ibm_cleveland and ibm_kobe, with up to 94 qubits active during the simulations. The workflow executed 9,200 quantum circuits over more than 100 hours and collected 1.3 times 10 to the ninth power individual measurements. On the classical side, Japan’s Fugaku and Miyabi-G supercomputers handled the bulk electronic structure computations that quantum hardware alone could not complete.
The hybrid method that tied these systems together is called EWF-TrimSQD, according to IBM’s announcement. It splits the problem so quantum processors handle the parts of the calculation where quantum effects are strongest, while classical machines manage the surrounding electronic environment. Nearly 6,000 quantum operations were executed within this framework. The approach allowed each technology to operate where it holds an advantage rather than forcing one platform to do everything.
For drug discovery teams, the practical takeaway is direct. Modeling how a drug candidate binds to a protein at the electronic level requires accounting for quantum mechanical interactions across thousands of atoms. Classical approximation methods lose accuracy at this scale, and brute-force quantum simulation of an entire protein remains far beyond current hardware. The hybrid split is what made the 12,635-atom calculation feasible, and it represents a template other research groups can study and adapt.
The work also illustrates how quantum computing is likely to enter pharmaceutical pipelines: not as a standalone replacement for existing infrastructure, but as a specialized module embedded in a larger high-performance computing environment. In this model, supercomputers handle tasks such as system preparation, basis-set selection, and large-scale density functional calculations, while quantum processors tackle smaller, high-precision subproblems that are classically intractable. The Cleveland Clinic–RIKEN–IBM collaboration effectively turned the quantum devices into accelerators for a specific slice of the electronic structure problem.
How the hybrid workflow changes molecular modeling
Electronic structure simulations at the scale of tens of thousands of atoms open up new categories of questions. Medicinal chemists care not only about whether a ligand can bind to a target pocket, but also about subtle effects such as charge redistribution, polarization of nearby residues, and the influence of water molecules and cofactors. These details can shift binding affinities by fractions of a kilocalorie per mole-small on an absolute scale, but large enough to change which compound becomes a viable lead.
Traditional techniques such as molecular mechanics and classical molecular dynamics approximate these effects with force fields tuned to experimental data. While effective for many tasks, they can miss quantum phenomena like charge transfer or multi-reference electronic states. By inserting quantum calculations into the most sensitive parts of a protein-ligand system, EWF-TrimSQD aims to capture these effects more faithfully without incurring the impossible cost of a fully quantum treatment.
The method’s design also hints at how future workflows might evolve as hardware improves. If quantum processors gain more qubits and lower error rates, the quantum region of the system could expand from a small active site to larger portions of the protein or solvent shell. Conversely, improved classical algorithms or better embedding schemes might reduce the number of quantum calls needed, optimizing cost and throughput. For now, the Cleveland Clinic–RIKEN–IBM result serves as a proof of principle that such a division of labor is technically viable at a size relevant to real drug targets.
What remains uncertain about the results
Several open questions surround the achievement. The raw measurement data and error-mitigation logs from the 1.3 billion shots have not been released beyond what appears in the arXiv preprint. Without access to those logs, outside researchers cannot yet evaluate how much noise correction was needed or how error rates may have affected final accuracy. Independent replication by a lab not affiliated with IBM, Cleveland Clinic, or RIKEN has not been reported.
The sources also do not include clinical outcome metrics or any wet-lab validation tied to the modeled protein-ligand complexes. A successful quantum chemistry simulation is not the same as a validated drug target. Whether the electronic structure data produced by EWF-TrimSQD is accurate enough to guide real medicinal chemistry decisions is a question the current publications leave open. Detailed runtime breakdowns comparing how much compute time Fugaku and Miyabi-G consumed versus the quantum processors have not been published in full either, making it difficult to assess the true efficiency gain of the hybrid approach over a purely classical alternative at smaller scales.
The claim that this is “the largest known to be simulated with quantum computers” comes from IBM’s own release. No independent benchmarking body has confirmed that characterization, though no competing claim at a higher atom count has surfaced in the scientific literature. Until peer-reviewed studies and third-party benchmarks appear, the result should be treated as a leading but still provisional record.
Separating primary evidence from promotional framing
Readers evaluating this development should distinguish between two layers of evidence. The first layer is the technical preprint, which provides specific numbers: 12,635 atoms, 156-qubit processors, 94 qubits used, 9,200 circuits, and 1.3 billion measurements. These are testable, falsifiable claims that other quantum computing groups can attempt to reproduce. The preprint is hosted on arXiv, a platform whose governance information is maintained by Cornell-affiliated staff, and while it has not yet undergone formal peer review, its metrics are concrete enough for the community to scrutinize.
The second layer is the promotional narrative. IBM’s press materials frame the result as evidence that quantum-centric supercomputing is ready to tackle problems classical machines “cannot touch.” That framing carries weight only if the accuracy of the quantum-assisted calculation matches or exceeds what classical methods achieve on smaller but comparable systems. The preprint does not yet provide that direct comparison in a way external reviewers have validated, nor does it present head-to-head benchmarks against state-of-the-art classical electronic structure packages at similar accuracy levels.
Another distinction involves the notion of “quantum advantage.” The Cleveland Clinic–RIKEN–IBM result demonstrates a form of scale advantage: a hybrid quantum-classical workflow has been run on a system size that classical quantum chemistry codes would struggle to treat at comparable theoretical rigor. But scale alone is not the only metric that matters. For pharmaceutical decision-making, what counts is predictive power-whether the method can correctly rank candidate molecules, anticipate off-target effects, or forecast resistance mutations. None of those application-level benchmarks appear in the current public documentation.
What to watch next
One useful signal to watch will be whether other pharmaceutical companies or academic labs adopt EWF-TrimSQD or similar hybrid workflows over the next few years. If the method proves reproducible, accurate, and cost-effective, it could trigger a measurable increase in collaborations between quantum computing providers and drug discovery teams, as well as a rise in patent filings related to quantum-assisted design. Conversely, if adoption remains limited, that may indicate unresolved issues with noise, scalability, or integration into existing computational chemistry pipelines.
Another indicator will be the appearance of peer-reviewed articles that benchmark hybrid quantum-classical simulations against gold-standard classical methods on smaller systems where exact or near-exact answers are known. Such studies would help clarify whether the additional complexity of incorporating quantum hardware delivers a tangible benefit in accuracy or efficiency for realistic drug discovery problems.
For now, the Cleveland Clinic–RIKEN–IBM collaboration marks a milestone: a demonstration that quantum processors can be woven into supercomputing workflows to tackle electronic structure problems at a scale that begins to mirror real biological targets. Whether this milestone becomes a foundation for routine pharmaceutical practice or remains a high-profile demonstration will depend on the next wave of data, validation, and independent replication.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
