Quentir Medicine Monitor
Evidence-based insights for quantum medicine. Published by Quentir Systems LLC · October 28, 2026.
Shionogi, the Osaka drug maker, took first place and a ¥40 million prize in the Quality of Life category of the NEDO Challenge on quantum computing, Japan's state-funded contest for quantum solutions to social problems. D-Wave Quantum announced the win on 28 October 2026 and tied it to a peer-reviewed study in which its quantum annealing hardware helped a generative AI model design drug-like molecules.
The study is the part a medicinal chemist can check. Published open access in Scientific Reports on 9 July 2026, Table 1 of the published article reports that 96.97 percent of the molecules generated with the annealer's help were chemically valid, against 61.95 percent for the same model with a classical sampler. The quantum-assisted model also produced a larger share of compounds with high drug-likeness scores than its training set contained, while repeating itself more often.
Those are computed scores on a public chemistry dataset. The study reports no synthesis and no biological testing of any generated molecule, and the prize money is meant to fund the next steps of that research.
What NEDO awarded on 15 October 2026, and what Shionogi's entry covered
NEDO, the New Energy and Industrial Technology Development Organization that operates under Japan's Ministry of Economy, Trade and Industry, held the award ceremony in Tokyo on 15 October 2026 and posted the official list of winners the following day. The contest ran in three areas: Society 5.0, Quality of Life and Cool Japan. First prize in each area carries ¥40 million, second ¥20 million and third ¥10 million. First place in Society 5.0 and Cool Japan was left vacant, which makes Shionogi the only first-place winner this season.
NEDO's list describes Shionogi's entry in Japanese as solving drug discovery problems through two complementary uses of quantum computers: generating the structures of candidate compounds, and evaluating the interaction energy between molecules. The other Quality of Life prizes went to two teams working on the gut and wider microbiome: Mori-san Daisuki Club in second place with a quantum search method for designing gut bacterial communities, and Q4Health in third place for microbiome design aimed at next-generation antimicrobial treatments.
The D-Wave study maps onto the first of Shionogi's two tracks. Quantum Computing Report wrote on 23 October 2026 that all nine general prize winners used IBM Quantum hardware, and it described Shionogi's entry as a quantum and high-performance computing hybrid that evaluates binding energy and generates structures through a method called Deep QSCI. D-Wave's announcement says the award covered Shionogi's broader research, including, among other achievements, the annealing study. A reader should therefore treat the prize as recognition for a two-part program run on more than one kind of quantum machine.
Quantum pillar: computing. Technology readiness: TRL 3 of 9. The method has run on a real D-Wave quantum annealer and produced better computed scores on a public chemistry dataset, and the study reports no generated molecule made in a laboratory or tested in cells, animals or patients.
How the Japan Tobacco and D-Wave team paired a generative model with the Advantage2 annealer
The eleven authors, led by Hayato Kunugi and with Masaru Tateno as corresponding author, worked at the Central Pharmaceutical Research Institute of Japan Tobacco in Takatsuki and at D-Wave Systems in Burnaby, Canada. Shionogi took over Japan Tobacco's pharmaceutical business through a company split it completed on 1 December 2025, which is why the paper carries Japan Tobacco affiliations and the prize went to Shionogi. Six of the eleven authors are D-Wave employees, and the paper declares no competing interests.
The model reads and writes molecules as SMILES strings, the line notation chemists use to type a structure. An encoder compresses each molecule into a short internal code and a decoder turns codes back into molecules. Between the two sits a Boltzmann machine, a statistical model that learns which codes are likely. Sampling from such a model is hard on a classical computer, and the team used D-Wave's Advantage2 annealer to draw those samples directly from its hardware.
The quantum hardware works with binary values while the neural network works with continuous ones. The authors bridge the two with what they call a neural hash function, which converts the network's signals into binary form and also keeps training stable. With classical sampling alone, this conversion raised validity from 52.2 percent with a standard technique to 62.0 percent. Training used a public subset of ChEMBL, the European Bioinformatics Institute's database of bioactive molecules, and an earlier model trained on the same subset reached 54 percent validity.
Which drug-likeness and validity results the authors report, and their limits
Validity here means that the open-source RDKit chemistry toolkit could convert a generated string into a molecular structure. Drug-likeness is measured with QED, a standard score from 0 to 1 that combines properties such as size, lipophilicity and polar surface area. Table 2 of the published article reports that 66.79 percent of the compounds from the annealer-assisted model were drug-like, with a QED above 0.7, against 43.15 percent for the strongest classical model and 31.61 percent in the training data. D-Wave's summary of 20 August 2026 gives the same three figures.
Two figures need care. The paper's text and D-Wave's announcement of 28 October 2026 compare 97 percent validity with 73 percent for the classical sampler, while Table 1 of the published article gives 61.95 percent for that classical model; 73 percent appears nowhere in that table. The table also shows a cost: only 51.92 percent of the valid molecules from the annealer-assisted model were unique, against 98.09 percent for the classical version, so the classical model ended with 6,077 distinct valid molecules out of 10,000 attempts and the annealer-assisted model with 5,035.
The authors stress that the model was not told to favor drug-like molecules. The classical sampler mostly reproduced the training distribution, while the annealer-assisted model shifted toward higher QED scores. The paper treats this as its central finding, because generative models tend to reproduce the quality of the examples they learn from.
Three limits matter for a pharmaceutical reader. QED is a calculated filter, and a high score says nothing about whether a compound binds its target, reaches the right tissue or is safe. The comparison runs against one classical sampler in one architecture, so it does not show that annealing beats every classical method. The code and generated molecules are available from the authors only on request, so other groups cannot rerun the comparison from public files. Quentir described a different hybrid design in July, where a quantum discriminator ran into a chemistry limit on validity; this study reports the opposite effect on validity, with a different architecture and real hardware.
What Shionogi's prize money funds next, and what hospitals and drug buyers can watch
D-Wave says the ¥40 million prize will support Shionogi's continued research into designing drug candidate molecules with quantum computing. Patients and hospital formularies will see no change from this award. Any compound found this way still faces the usual sequence of synthesis, laboratory assays, animal studies and clinical trials, which takes many years.
Three kinds of evidence would move the work forward. One is synthesis and assay data for molecules generated with the annealer's help, against a defined disease target. Another is a comparison with strong classical generators trained on the same data. A third is published results from Shionogi's second track, the binding-energy calculations, showing whether they change which candidates chemists choose to make.
Sources
Primary source: the Scientific Reports study of 9 July 2026 by Hayato Kunugi, Mohsen Rahmani, Yosuke Iyama, Yutaro Hirono, Akira Suma, Matthew Woolway, Vladimir Vargas-Calderón, William Kim, Kevin Chern, Mohammad Amin and Masaru Tateno of Japan Tobacco's Central Pharmaceutical Research Institute and D-Wave Systems. Also drawn on: NEDO's announcement of the winners of 16 October 2026, D-Wave's announcement of 28 October 2026 and its blog summary of 20 August 2026, Quantum Computing Report's coverage of 23 October 2026, Shionogi's notice of 1 December 2025, and earlier Quentir coverage.