---
title: "Crossing the 12,000-Atom Barrier: How Cleveland Clinic, RIKEN and IBM Simulated a Protein on Quantum Hardware"
slug: quantum-protein-simulation-12635-atoms-cleveland-clinic-riken-ibm
category: tech
category_label: "Tech"
author: "BrainWavePost Staff"
date: 2026-09-10
tags: ["quantum computing", "IBM", "RIKEN", "Cleveland Clinic", "drug discovery", "supercomputing", "Fugaku", "quantum chemistry"]
read_time_minutes: 9
canonical_url: https://brainwavepost.com/article/quantum-protein-simulation-12635-atoms-cleveland-clinic-riken-ibm
source: BrainWavePost
---

# Crossing the 12,000-Atom Barrier: How Cleveland Clinic, RIKEN and IBM Simulated a Protein on Quantum Hardware

*Tech · 2026-09-10 · BrainWavePost Staff · 9 min read*

> A joint team used two 156-qubit IBM Heron processors and three supercomputers to compute the electronic structure of protein-ligand complexes up to 12,635 atoms — the largest biologically meaningful molecules known to be modelled with quantum hardware.

> **How this article is sourced** _(info)_
>
> Claims here come from the Cleveland Clinic / IBM / RIKEN press release (5 May 2026), RIKEN's own announcement, IBM's technical blog, and the arXiv preprint 'Crossing the 12,000-atom barrier with heterogeneous quantum-classical supercomputing' (Merz et al., v2, 5 September 2026). Where a number is a vendor claim rather than an independently reproduced result, this article says so. [1][2][3][4]

## What was actually announced

On 5 May 2026, Cleveland Clinic, Japan's RIKEN and IBM announced that they had simulated the electronic structure of two protein-ligand complexes spanning 11,608 and 12,635 atoms — described as the largest known simulations of biologically meaningful molecules performed with quantum hardware. [1][2][4]

The larger of the two systems is a complex built around trypsin, a digestive enzyme long used as a workhorse in structural biology. [1] The point of the exercise was not the enzyme itself but the size threshold: quantum chemistry calculations on quantum processors had previously been confined to small molecules, and the team reports roughly a 40-fold jump in system size over what the same method could handle six months earlier. [1][4]

The work is reported in a preprint on arXiv rather than in a peer-reviewed journal, a distinction worth keeping in mind. [1][4] It builds on earlier steps from the same collaboration, including a 303-atom simulation of Trp-cage, a compact 20-amino-acid mini-protein. [1]

## The method: divide, then delegate

No quantum computer today can represent a 12,635-atom molecule directly. The approach, which IBM calls quantum-centric supercomputing, splits the problem so each machine does what it is good at. [3][4]

1. Classical computers decompose the protein-ligand complex into smaller fragments using quantum embedding, so only chemically critical regions need quantum treatment. [3][4]
2. Quantum processors sample the electronic configurations of those fragments — the part where classical approximation is weakest. [3][4]
3. Supercomputers then solve for the fragment wavefunctions via optimised subspace diagonalisation and reassemble the pieces into a description of the whole molecule. [3][4]

The hybrid algorithm behind the jump is named EWF-TrimSQD in the announcement; the team credits it with cutting computational overhead enough to make the larger systems tractable. [1][4]

## The hardware bill

The scale of the run is unusually well documented for a quantum-computing milestone. Sampling ran on two 156-qubit IBM Quantum Heron processors — ibm_cleveland, installed at Cleveland Clinic in Ohio, and ibm_kobe at RIKEN — using up to 94 qubits, executing 21,006 circuits over more than 239 hours and collecting about 3.0 billion measurement outcomes. The authors describe it as the most resource-intensive heterogeneous quantum-classical computation for quantum chemistry to date. [1][4]

The classical half ran on Fugaku at RIKEN, Miyabi-G (operated by the University of Tokyo and the University of Tsukuba) and ROQUO, with the distributed linear-algebra kernels reported at 72.5% parallel efficiency. [1][4]

- **12,635** — Atoms in the largest complex simulated [1][4]
- **94** — Qubits used, on 156-qubit Heron processors [1][4]
- **239 h** — Quantum processor runtime across 21,006 circuits [4]
- **72.5%** — Parallel efficiency of the classical diagonalisation step [4]

## Why protein-ligand binding is the target

Predicting how tightly a candidate drug molecule binds to a protein is one of the most consequential and least reliable calculations in early drug discovery. Existing computational methods trade accuracy for tractability as molecules grow, and getting binding energies wrong early means expensive failures later in a development cycle that can stretch beyond a decade. [1]

The two complexes chosen span different physical regimes — one dominated by dispersion forces, the other by electrostatics — which is a deliberate test of whether the workflow generalises rather than fitting one convenient case. [4] The team also reports the first heterogeneous quantum-classical protein-ligand binding prediction using a mixed-basis set, delivered through an automated end-to-end workflow. [4]

> By crossing the 12,000-atom barrier, we have significantly expanded the scale of biologically meaningful molecular simulations possible with quantum computing and demonstrated a framework for applying these methods to scientifically relevant problems at a larger scale.
>
> — Kenneth Merz, Ph.D., lead author, Cleveland Clinic Computational Life Sciences [1]

## How accurate is it?

The most meaningful accuracy claim is not the atom count but the benchmark: the hybrid method's fragment energies matched coupled-cluster (CCSD) accuracy, CCSD being a well-established classical reference standard in quantum chemistry. [4] The team also reports up to a 210-fold accuracy improvement in a key workflow step relative to its own prior state of the art six months earlier. [1][4]

That framing matters. This is not a demonstration that quantum hardware beat classical computing at a task classical computing cannot do. It is a demonstration that a quantum-classical workflow reached trusted classical accuracy on fragments of a molecule far larger than previously attempted, using an architecture that could scale further. [3][4]

## The accuracy box: what is solid and what is not

- Solid: two protein-ligand complexes of 11,608 and 12,635 atoms had their electronic structure computed through a quantum-classical workflow, on documented hardware with documented circuit counts. [1][4]
- Solid: fragment energies matched CCSD accuracy, and the classical diagonalisation scaled across three supercomputers. [4]
- Solid: the project is supported by Japan's NEDO under METI, within a programme led by Mitsuhisa Sato. [1]
- Vendor-framed: the '40x larger' and 'up to 210x more accurate' figures are comparisons against the same team's earlier results, not against the best available classical methods. [1][4]
- Careful with the framing: 'unprecedented accuracy' for predicting molecular interactions overstates the present state. Binding prediction was demonstrated, not validated against experimental affinities at scale. [4]
- Not established: any quantum advantage for drug discovery. Classical methods remain the practical tool; this is a scaling milestone on a path toward usefulness. [3][4]
- Status: a preprint on arXiv (v2, 5 September 2026), not yet a peer-reviewed journal publication. [4]

## What it does not mean

It does not mean a quantum computer designed a drug. Nothing here produced a therapeutic candidate, and no patient-facing claim follows from it. The calculation describes electronic structure and a binding prediction for known complexes — a step in a pipeline, not a discovery. [1][4]

It also does not mean molecular dynamics has been solved. IBM's own framing separates two capabilities: modelling how atoms move as biological processes unfold, and accurately computing their energies. This work provides evidence for the second, not the first. [1]

## What to watch next

1. Peer review of the preprint, and whether the CCSD-accuracy comparison holds under scrutiny. [4]
2. Whether predicted binding energies get validated against measured experimental affinities. [4]
3. Whether the workflow extends to enzyme catalysis and drug mechanisms that today can only be studied experimentally. [1]
4. Whether other groups reproduce the approach on different hardware, which is what turns a vendor milestone into a method. [3]

The collaboration's work has since advanced to the finals for the 2026 ACM Gordon Bell Prize, a recognition that judges high-performance computing achievements rather than quantum hype. [1]

## Sources and further reading

- [1] IBM Newsroom — 'Cleveland Clinic, RIKEN, and IBM Model a 12,635-Atom Protein – the Largest Known to Be Simulated with Quantum Computers' (5 May 2026): https://newsroom.ibm.com/2026-05-05-cleveland-clinic,-riken,-and-ibm-model-a-12,635-atom-protein-the-largest-known-to-be-simulated-with-quantum-computers
- [2] RIKEN — 'Collaboration with Cleveland Clinic and IBM to model a 12,635-atom protein using quantum computers' (5 May 2026): https://www.riken.jp/en/news_pubs/news/2026/20260505_1/index.html
- [3] IBM Quantum — 'Researchers use quantum-centric supercomputing to simulate 12,635-atom protein complex' (technical blog, 5 May 2026): https://www.ibm.com/quantum/blog/cleveland-clinic-riken-chemistry
- [4] Merz et al. — 'Crossing the 12,000-atom barrier with heterogeneous quantum-classical supercomputing: quantum chemistry of protein-ligand complexes', arXiv:2605.01138 (v2, 5 September 2026), DOI 10.48550/arXiv.2605.01138: https://arxiv.org/abs/2605.01138
- [5] Cleveland Clinic Newsroom — press release on the 12,635-atom protein simulation (5 May 2026): https://newsroom.clevelandclinic.org/2026/05/05/cleveland-clinic-riken-and-ibm-model-a-12635-atom-protein--the-largest-known-to-be-simulated-with-quantum-computers

---

_Canonical article: [https://brainwavepost.com/article/quantum-protein-simulation-12635-atoms-cleveland-clinic-riken-ibm](https://brainwavepost.com/article/quantum-protein-simulation-12635-atoms-cleveland-clinic-riken-ibm) — © BrainWavePost. Educational content; see the article page for full disclaimers._