Quentir Medicine Monitor
Evidence-based insights for quantum medicine. Published by Quentir Systems LLC · October 26, 2026.
Daiichi Sankyo is using software from the Boston start-up QSimulate to bring quantum chemistry into one of the most routine and most expensive judgments in drug discovery: how tightly a candidate molecule will bind to its protein target. The main public account is a Japanese-language blog on the Amazon Web Services site, published on 11 June 2026.
That blog was written jointly by scientists from Daiichi Sankyo's Smart Research Institute II, by QSimulate, and by an AWS solutions architect, and it describes how the company evaluated and then adopted QSimulate's QM-FEP method, a version of free energy perturbation that uses quantum mechanics for the energy of the interaction between drug and protein.
Predicting binding affinity well matters because it decides which molecules a medicinal chemistry team actually makes. Each round of the design, make, test and analyze cycle costs weeks of synthesis and assay work. A computer model that ranks candidates more reliably lets chemists spend that time on the compounds most likely to work, and lets them drop weak ideas before anyone orders reagents.
No quantum computer is involved here. QSimulate's software runs quantum mechanical calculations on ordinary graphics processors in the cloud. The interest for medicine lies in a large pharmaceutical company putting physics-level chemistry into its everyday design work today, years before quantum hardware is expected to take on the same problem.
Why Daiichi Sankyo wanted quantum mechanics in its free energy calculations
The Daiichi Sankyo authors explain the problem in practical terms. Their teams already use structure-based drug design and conventional free energy perturbation, which estimates binding by simulating the gradual transformation of one molecule into another inside the protein. That method depends on a classical force field, a fixed set of rules for how atoms push and pull on each other. The authors write that more accurate energy functions are needed, especially for the chemical modalities they expect to grow more complex, where conventional FEP has struggled with non-classical interactions and with covalent drugs that form a chemical bond with their target.
QSimulate's product for this work is called QUELO. According to the blog, it combines a quantum mechanical force field with the company's own AI-trained force field, and it extends accurate affinity prediction to ligands that coordinate a metal ion and to covalent systems. QSimulate describes QUELO as the only commercial FEP product based on quantum mechanics. The company was founded in Boston by its chief executive Toru Shiozaki, formerly a chemistry professor at Northwestern University, and develops the software with teams in Boston, Berkeley, Belgium and Tokyo.
The speed figures are the company's own. The blog states that a mixed single- and double-precision algorithm running on Amazon EC2 GPU instances makes the quantum chemical step about 1,000 times faster than conventional quantum chemistry, so that one snapshot of a protein and ligand complex is processed in milliseconds. Simulations that used to take months now take hours, and computing cost falls to between one hundredth and one thousandth, according to the same text.
Quantum pillar: simulation. Technology readiness: TRL 7 of 9. The quantum chemistry software is already running inside Daiichi Sankyo's real drug discovery projects, but the only account of its accuracy comes from the company and its supplier, with no published comparison figures, so routine proven use has not yet been shown in public.
What Daiichi Sankyo reports about accuracy, and what its account leaves out
The results section is brief. The Daiichi Sankyo authors report good prediction accuracy from calculations with the quantum mechanical force field, and they say that adding the AI force field kept that accuracy while shortening the calculation. The quantum mechanical version handled the non-classical interactions that their earlier methods could not. They also found that running ordinary jobs on cheaper spot instances, and urgent ones on on-demand instances, gave the research teams the flexibility they needed.
The strongest statement is about use. The authors write that the method is already applied across several drug discovery programs and targets, and that in an actual project it contributed to the creation of a new chemotype, a structurally new family of candidate molecules. They expect further gains for covalent drugs, where chemists must compare the reactive groups, called warheads, that form the bond with the protein.
The Daiichi Sankyo account gives no numbers for that accuracy: no correlation with measured affinities, no error in kilocalories per mole, no comparison set of compounds, and no description of the chemotype or its target. A reader therefore has the company's judgment that the tool works, published on a cloud vendor's site with the supplier as co-author. That is useful information about adoption, and it falls short of independent evidence of how well the predictions hold.
What a 2020 MCL1 preprint and Otsuka's August 2026 adoption add
Earlier public work gives the method some scientific footing. In a 2020 preprint, Narbe Mardirossian, Yuhang Wang, David Pearlman, Garnet Kin-Lic Chan and Toru Shiozaki described algorithms that run full density functional theory on protein and ligand complexes of about 2,500 atoms in tens of minutes on cloud hardware. For a test set of 22 ligands binding to the cancer target MCL1, density functional scoring outperformed classical free energy perturbation. That paper studied an earlier scoring method, which is related to the present QUELO QM-FEP product without being the same thing. Several of the authors worked at or held stakes in QSimulate, and one protein system is a narrow test, so the result supports plausibility more than general superiority.
Method details decide whether quantum chemistry helps or misleads. Quentir looked at a quantum-classical calculation by Cleveland Clinic, RIKEN and IBM in which a minimal basis set gave the electronic binding energy of T4 lysozyme and its ligand the wrong sign, a result the team corrected by using richer orbitals near the ligand together with a tighter bath threshold, without separating the effect of each change. That calculation concerned an electronic energy, which is a different quantity from the binding free energy that FEP estimates. The broader point carries over: a quantum mechanical label on a calculation says little until the level of theory and the validation set are shown.
Daiichi Sankyo is also not alone. QSimulate announced on 27 August 2026 that Otsuka Pharmaceutical had deployed QUELO as part of its research digital transformation, again pointing to covalent inhibitors and metal-coordinated drug molecules. With the Daiichi Sankyo account in June and the Otsuka announcement in August, two large Japanese drug makers have now said publicly that they use the tool.
What clinicians, hospital research leaders and pharma buyers can watch for in 2027
For patients, any benefit arrives slowly and indirectly. Better affinity prediction can shorten early discovery and widen the range of drug types chemists can design, including covalent inhibitors used in oncology. Affinity predictions do not establish whether a drug is safe or effective in patients, and a new chemotype from 2026 would need years of preclinical and clinical work before it reached a hospital formulary.
The useful signals are concrete. One is a peer-reviewed or conference paper from Daiichi Sankyo or Otsuka with prospective predictions checked against measured affinities. Another is a candidate whose design history credits quantum mechanical scoring when it enters clinical development. A third is a published comparison of QM-FEP against classical FEP on covalent or metal-binding series, the cases where the method claims its advantage. Until such data appear, the AWS blog by the Daiichi Sankyo and QSimulate authors remains the main public account. A Nikkei Tech Foresight article dated 23 October 2026 on its page, of which only the subscription preview is public, describes the same collaboration as an effort to use quantum chemical calculation for predicting compound properties and improving affinity prediction.
Sources
Primary source: the AWS Japan blog of 11 June 2026 by Akitoshi Okada, Ayako Moritomo and Takayuki Serizawa of Daiichi Sankyo, Toru Shiozaki and Sho Imoto of QSimulate, and Takehiro Nakajima of AWS. Also drawn on: the public preview of Nikkei Tech Foresight's article dated 23 October 2026, QSimulate's Otsuka Pharmaceutical announcement of 27 August 2026, the 2020 preprint by Mardirossian, Wang, Pearlman, Chan and Shiozaki, and earlier Quentir coverage.