Aqemia

Qemi produces everything Aqemia sells and everything it develops. Physics algorithms predict how strongly candidate molecules bind a target, those predictions train a generative model, and the model proposes molecules tailored to that target. Partners such as Sanofi receive designs against targets they nominate, and the same engine starts each of the company's own programs.

With no laboratory or clinical operation of its own, Aqemia has nothing to sell apart from what the platform designs.

Contract laboratories and partner scientists make and test Aqemia's designs before any of them moves forward. On partnered programs Aqemia designs molecules against targets the partner chooses, and Sanofi leads the wet lab research, development and commercialization, with Aqemia researchers in Paris and London working alongside Sanofi researchers in Boston, Frankfurt and Paris. On its own programs, contract research organizations run the experiments. Servier's first engagement was a blind test that Servier scored before it signed.

How Aqemia's chemists review or reject what the generative model proposes is not described.

Qemi has two layers. Statistical and quantum mechanics algorithms compute the binding free energy between a molecule and its target, and those results train a generative model that proposes new molecules. The company attributes its speed, physics level accuracy up to 10,000 times faster than conventional methods, to new mathematics from quantum mechanics rather than computing power, built over 12 years of academic research at CNRS, the École Normale Supérieure, Oxford and Cambridge that produced more than 40 peer reviewed publications. A 2021 preprint written with Sanofi scientists applied absolute binding free energy calculations to reposition approved molecules against the SARS-CoV-2 main protease.

No architecture, training method or accuracy benchmark for Qemi itself is public.

Aqemia builds both layers of Qemi itself: physics algorithms that grew out of Levesque's academic research in quantum and statistical mechanics, and a generative model trained on the data those algorithms compute. The company says the approach removes the need for experimental training data. Laboratory work on its own programs goes to contract research organizations.

The company lists no compute or hosting provider, open source components or subprocessors.

Partners have tested Aqemia's designs in their own programs since 2020. Servier ran a blind test of the technology in early 2021, signed an immuno oncology collaboration that December, renewed it in 2023 against a target it described as undruggable, and reported in June 2024 that a series of designed molecules was active in vitro and in vivo. Sanofi began with COVID-19 treatments in 2020, extended that first collaboration into oncology in 2022, signed a multi year discovery agreement in December 2023, and in July 2026 nominated a new target that triggered a milestone payment. Janssen ran a pilot study in 2022. On its own side, the company reports three candidates in preclinical development across more than a dozen internal programs, and says its wholly owned programs typically show superior efficacy without toxicity in vivo.

The targets, molecules and assay data behind these results have not been made public.

Qemi's inputs are targets, molecular structures and computed binding energies, so no patient health information is involved. The company describes the physics layer as generating the data it needs in house from the start of each program instead of training on large experimental data sets, and it also says Qemi keeps learning from every program it runs. On its website, newsletter and other prospect data is kept for at most three years after the last contact and shared with its web host and the marketing subcontractors that run its newsletter.

How targets and designs from one partner are kept apart from work for another is not described.

Aqemia works with molecular structures, target information and computed binding energies rather than patient records, as a French company subject to European data protection law. Its website legal notice, governed by French law, commits to the French data protection act of 1978 and to European Regulation 2016/679, and covers the names and email addresses visitors submit through contact forms and newsletter sign ups.

The company makes no statement on HIPAA or on business associate status.

Partners such as Sanofi entrust Aqemia with the targets they nominate, and the company holds the molecules it designs against them, material drug companies treat as core intellectual property. Aqemia's researchers work on that material from offices in Paris and London, and contract research organizations receive the designs they make and test for its own programs.

The company names no security certification, trust center or security program.

Aqemia designs drug candidates at the discovery stage and sells no software, so Qemi sits outside device regulation. Its molecules would follow the drug pathway: the Sanofi agreement covers first hits through development candidate selection, and Sanofi carries partnered programs into development after that.

None of the three preclinical candidates has a named investigational new drug filing or trial start.

Qemi's physics first design is the company's response to the limits of models trained on past drugs. It argues that AI trained on past drugs struggles with new targets, so Qemi predicts how a future drug interacts with its target atom by atom rather than by resemblance to past cases, and the company describes its generative model as free from bias and tailored to each target. Qemi also designs molecules to stay selective against targets a drug must avoid.

The company has published no study of where Qemi's predictions depart from measured results, and no model card or governance policy.

A design moves toward patients only after partner or contract laboratories have made and tested it, and partners carry the programs they take forward into development. On partnered work Sanofi picks the targets, and the risk on each program sits inside collaboration agreements between the companies.

The company publishes no accuracy figure for Qemi's predictions and no remedy for a design that fails.

The customers are drug discovery teams rather than care providers, so the work never touches an electronic health record or a clinical workflow. Designs pass from Qemi to partner scientists or to contract research organizations for laboratory work, and partners receive molecules rather than software to install or connect.

The systems and data formats used for those handoffs are not described.

Qemi is operated by Aqemia's own research teams in Paris and London, and partners receive designs rather than a hosted service. The website legal notice says personal data is processed within the European Union, while allowing that technical subcontractors such as Google in the United States may receive it, in which case they must be bound by clauses comparable to the European Commission's or by binding corporate rules.

That notice covers website data only, and no hosting location is given for the targets and designs partners entrust to the platform.

Negotiated discovery partnerships are how Aqemia sells its work, with partners paying for designed molecules rather than software. The Sanofi agreement announced in December 2023 carries the published figures: eligibility for up to $140 million in upfront and milestone payments across programs, for work running from the first hits to the choice of a development candidate, with Sanofi responsible for wet lab research, development and commercialization. That sum is a ceiling across every program, milestones included. The July 2026 expansion added a target and triggered a further payment.

The size of that payment and the terms of the Servier and Janssen work are not public.

The work is small molecule design at the discovery stage, from first hits to development candidates, across oncology and immuno oncology, immunology, inflammation and central nervous system disease. A 2025 France 2030 grant extended it to small molecules that target RNA and to epitranscriptomics. Programs run two ways: broad exploration that scans many targets for the most promising opportunities, and indication first programs, which the company calls Moonshots, that pursue one disease of high unmet need through several targets at once. Clinical development belongs to partners, or to the biotech spin offs the company plans for its own candidates.

No program names its target or molecule.