Born in 2020, deeptech Qubit Pharmaceuticals has since been dedicated to developing new, safer and more effective drug candidates. To do this, it is also working, in parallel, on more fundamental research work, focused in particular on quantum computing. An area in which it recently made major progress, as revealed by its CEO and its scientific director during a press conference organized this summer.
« We are moving, step by step, towards quantum computing. » This is the observation presented to us, at the beginning of 2021, by Robert Marino, CEO of the young company Qubit Pharmaceuticals. Since its creation in 2020, the Parisian deeptech has focused its efforts around simulation and molecular modeling accelerated by hybrid computing – an approach combining high performance computing (HPC[1]) and quantum computing – with a central objective: discovering new drugs.
Faced with this prospect of growth in quantum computing that its director mentioned, the start-up has, since its beginnings, chosen to take the lead… “We need todaytoday prepare for the quantum transition, to be able to meet the needs of our customers, when the time comes », explained to us, in fact, still in 2021, the same Robert Marino.
Nearly four years later, it is clear that Qubit Pharmaceuticals got it right, as evidenced by the major advances it has made and on which it lifted the veil last summer, as part of a conference of press organized within the Paris Biotech Santé Cochin Nursery which hosts it. “ Quite exceptional results, which now allowconsider, in the short term, theuse of thequantum computing in production », as announced, in the preamble of this summer meeting with the press, Robert Marino.
Emulation: qubits before qubits
In addition to its applied research work, focused on the co-design of therapeutic molecules alongside CRO[2] French (read in box below), the company which today has around sixty employees is pursuing a major more fundamental research program in the field of high-performance computing and quantum computing, and has its own cluster calculation center, located north of Paris. A particularly powerful tool – composed of more than 200 GPUs[3] – which it uses both for its needs in terms of high-performance computing and training of AI algorithms, but also for an application directly linked to quantum computing: qubit emulation.
« It is indeed possible to emulate a quantum computer on a classical computer », schematizes Jean-Philip Piquemal, Professor at Sorbonne University, co-founder and scientific director of Qubit Pharmaceuticals. A possibility which has led deeptech to develop its own emulator: Hyperion-1. “ It seemed important to us not to wait and to start now to develop the software that will run, tomorrow, on real quantum computers. », underlines Jean-Philip Piquemal.
Thanks to this tool, the company has notably succeeded, as it announced at the end of last year, in emulating 40 so-called “logical” qubits, these assemblies of physical qubits making it possible to correct errors – and they are numerous – arising in particular from the phenomenon of quantum decoherence.
40 logical qubits… Still very far from the approximately 250 expected to be necessary to accurately solve the Schrödinger equation and thus succeed in determining, as faithfully as by experiment, certain properties of small molecules such as dihydrogen (H2), or lithium hydride (LiH). “ This is called chemical precision: achieving a calculation result indistinguishable from experiment », explains Jean-Philip Piquemal.
Hybrid computing, a winning bet
And yet, Qubit Pharmaceuticals succeeded, as it revealed during this meeting with the press organized last July. And this, moreover, without even exploiting the full potential of its Hyperion-1 emulator: around twenty qubits were in fact enough for it – around thirty, at most, for certain molecules – as its scientific director explains to us: “ Everyone thought that it would take hundreds, even thousands or tens of thousands of qubits… With Hyperion’s hybrid technology – which allows you to sort out the different sub-parts of a complex calculation. and to direct them, depending on the interest that this may have in terms of speed of execution, either towards HPC or towards quantum computing – we have however shown that this is possible with ten times fewer qubits than expected “. A sort of “shortcut” on the path that leads to chemical precision via quantum calculation, to use the title of a pre-publication describing, in detail, these results and the method used by researchers at Qubit Pharmaceuticals and Sorbonne University to achieve this. “ However, if reaching 200 qubits in a machine remains a real challenge, it is less so for around twenty! », underlines Jean-Philip Piquemal.
This breakthrough has enabled deeptech to obtain funding of 8 million euros as part of the France 2030 plan. Enough to continue the development of Hyperion-1 and thus continue to chart the path to solutions that will turn , tomorrow, on real quantum computers. “ Qubit emulation allows us to not have to wait for the availability of these “physical” machines to develop our algorithms. », Indeed reminds Jean-Philip Piquemal again. But that’s not all: as the Sorbonne University Professor concludes, “this also allows us to participate, alongside their manufacturers, in the design of these future machines, by defining now the number of qubits that we will really need ».
Computer-assisted design of therapeutic molecules: a rapidly accelerating approach
Over the last decade, the use of tools for computer-assisted design of therapeutic molecules – based, in particular, on AI algorithms – has experienced real growth, as evidenced by the CEO of Qubit Pharmaceuticals , Robert Marino: “ Today, nearly 10% of molecules undergoing clinical trials have been discovered using AI tools and, more broadly, computer-aided design. ».
A rise in power accompanied by a change of approach: used yesterday a posteriori to understand the effect – or absence of effect – of a molecule based on experimental data, these tools are today more used a a priori, before the syntheses.
Although the way in which it is implemented is therefore changing significantly, the computer-assisted design of therapeutic molecules remains, despite everything, quite largely centered around traditional targets, starting with proteins, such as kinases. Biomolecules for which experimental data – essential for training AI algorithms – are indeed abundant. A state of affairs that Qubit Pharmaceuticals has decided to take on the wrong foot… “ Our positioning consists, in fact, of seeking new targets – notably RNA – for which we currently have very little data, while there is a strong unsatisfied medical need.expose Robert Marino. To do this, we have developed methods based on three very complementary technological pillars – high-performance computing, AI, and quantum computing – which allow us, on the one hand, to simulate and validate molecules, but also, on the other hand, to generate new ones. »
Enough to reduce, according to the manager, by a factor of 10 to 20 the number of syntheses and biological tests necessary for the validation of new therapeutic molecules. Drug candidates which, as part of the work carried out by Qubit Pharmaceuticals, are mainly intended, for the moment, for the treatment of certain cancers and chronic inflammatory diseases.
[1] Pour high performance computing.
[2] Contract research organization, or contract research company: company specializing in the provision of services in the field of biomedical research intended, in particular, for the pharmaceutical industry.
[3] Graphics Processing Unit: graphics processor.
