Without DGX, we wouldn’t have the level of expertise we do today and MAbSilico wouldn’t have been able to grow as quickly. We can now solve the issues confronting pharmaceutical company biologists, offering them solutions against the backdrop of our now vast training capacities…
Dr Thomas BOURQUARD, Co-founder and Computer Science CSO, MAbSilico.
|
PROFILE.
After completing his thesis at the Paris IX Dauphine University’s Institute for Molecular and Cellular Biology Research (IBBMC), Thomas BOURQUARD dedicated himself to various computer science research projects at Nancy’s INRIA and Tours’ INRA research institutes. There he worked alongside Vincent PUARD, CEO MAbSilico; Astrid MUSNIER, Head of Biology MAbSilico and Anne Poupon, CTO MAbSilico. Thomas Bourquard then travelled to the United States where he stayed for just over four years working on integrating AI in genome studies in infectious and neurodegenerative diseases. MAbSilico was founded in parallel to his research as the result of a project triggered by his work with Astrid MUSNIER. The company’s development was made tangible notably thanks to the support of partnerships with academics, biotechnology companies, CHRU Tours university hospital, and others. |
MAbSilico(1) is a French deep tech company founded in 2017 by four researchers. They design and implement IT solutions predominantly for discovering and developing biomedicine and antibodies.
Their concept is to provide integrated AI technology to discover antibodies—and they’re the world leaders in it. This approach minimises the risk of failure to develop medication, which is generally 95%, in addition to shortening discovery and pre-clinic phases, all while reinforcing the protection of intellectual property.
Since its creation, MAbSilico has participated in 10 research programmes conducted in France, Europe, and in the United States, with a view to fighting cancer, infectious diseases and modulating reproduction. It also boasts more than fifty customers the world over ranging from startups to pharmaceutical laboratories.
They’re currently located on two sites and are planning a third in Paris for 2023. A project made feasible thanks to support from pharmaceutical company NOVO NORDISK, enabling them to use the Hotel Dieu Biolabs laboratories and offices, including systems such as the DGX NVIDIA A100 supplied by Bechtle.
MAbSilico is always either pursuing their own specific therapeutic goal or acting on behalf of a client and this is what spurs them on to design new digital sequences and develop new antibodies.
|
Antibody key figures
|
The aim of the IT is therefore to reduce the time, cost, and above all the 95% failure rate of developing medication. Parameters that are “crucial for patients’ wellbeing and ensure that our investments result in success,” says Vincent PUARD, CEO MAbSilico.
Certain steps typical in discovering antibodies were no longer feasible without employing a solution as performant as the HPC system with extensive AI analysis functions
such as the DGX NVIDIA A100. A therapeutic goal or an antibody is a series of letters that needs to be reproduced as a 3D model. Once this has been completed, we can then link the therapeutic goal with an antibody to see how they integrate with one another. At first glance this sounds easy, and with a limited number of units it can be, but not when you’re dealing with numbers in the millions.
|
This project therefore posed the following challenges:
|
SOLUTION.
The DGX NVIDIA A100 system is an AI-ready data centre designed with the aim of providing unparalleled processing power. It offers striking, data-centre-level performance without the need for an additional IT infrastructure—an innovative solution that allows MAbSilico to accelerate all of its projects and obtain optimum results more quickly.
Solution characteristics:
- 2.5 petaFLOPS of unlimited performance
- 64-CORE AMD CPU and PCIE GEN4, 3.2x more cores to power more users and intensive AI.
- 512 GB of system memory.
- Cooling system.
- A world-class AI platform that doesn’t require complex installation or special IT support.
- The only system in the world with four fully connected NVIFDIA A100 T processors and can be expanded up to 320 GB GPU memory.
- 7.68 terrabytes (TB) PCIE GEN4. NVME SOLID-STATE DRIVE (SSD). It offers storage system performance of 1.4 M IOPS, twice as fast as the NVMe PCIe Gen3 SSD.
Since we started using the system, we’ve already gathered four terrabytes of data and can now train new models to respond to the needs of the biopharmaceutical industry.
Thomas BOURQUARD, CSO Computer science, MAbSilico.
The standard today is the GPU(4)with 2 to 3 million parameters to train. The topic of memory is therefore important to MAbSilico. They were able to achieve up to between 80% and 93% correct predictions in detecting the right zones for discovering an epitope (antigen) by a paratope (antibody) with the solution thanks to NDIVIDA and the Inception program. They had met with several companies offering suitable solutions, but following efficient and quick first contact, Bechtle was selected to provide support throughout the project.
According to Vincent PUARD, CEO MAbSilico, a key factor in their choice was that Bechtle gave them the opportunity to test the device over several weeks to see whether it corresponded to their needs before making a concrete commitment.
The initial contract also includes a support element—a huge benefit allowing MAbSilico access to the top experts.
(1) MAbSilico – A portmanteau of MAb (a monoclonal antibody) and Silico (IT models).
(2) Therapeutic antibodies –Also known as monoclonal antibodies (MAbs) and their derivatives, these represent the majority of therapeutic proteins currently used in the fight against cancer and belong to the group referred to as biomedicine.
(3) Scalability – An IT system (or one of its components) that is able to adapt in size.
(4) GPU - Also called graphics coprocessor in certain systems, it is a processing unit that can be present in the form of an integrated circuit on a graphics card or motherboard.