Executives from French AI drug discovery unicorn Owkin are launching a startup that will develop a “large-language model” (LLM) — the same technology powering ChatGPT — for biotechnology, according to four sources with direct knowledge of the matter.
Sifted understands that the company, Bioptimus, was looking to raise a €20m seed round and one source involved with the fundraising says that it has secured funding above that target. The final figure could be different, however, once the company has closed the fundraising round.
The company’s registration documents state that Bioptimus will be led by Owkin’s current chief R&D officer and ex-Google researcher Jean-Philippe Vert, together with senior Owkin executives David Cahané and Eric Durand.
Owkin is also a signatory to the company’s registration documents, and Bioptimus is HQ'd at the same address as Owkin in Paris.
Owkin and Bioptimus declined to comment.
LLMs, which can imitate human language and generate new content such as images, sounds or software, are AI models trained on huge datasets. Investors and tech entrepreneurs believe this tech has the potential to significantly boost productivity when applied to specific industries.
Owkin uses another kind of AI, predictive AI — models trained to identify patterns from existing data to make predictions about future events — to understand and improve the effect of drugs and treatments on different patient groups. It has partnered with global pharmaceutical companies such as Sanofi, Bristol-Myers Squibb and MSD.
LLMs offer many potential ways to supplement Owkin’s efforts to make drug discovery faster and cheaper.
An LLM for biotech
Bioptimus will build a “foundational model for biology that will transform multiscale data into actionable representations to fuel breakthrough discoveries,” according to its website. LLMs are a type of foundational model.
Multiscale data in biology refers to the type of data used to represent organisms, which exist on multiple scales — cells, atoms, molecules, tissues, organs and so on.
Simulating the growth or development of an organism, for example its reaction to a drug, is very complex because it requires accounting for a myriad of changes and interactions, all of them data points, at various different scales.
While traditional AI models are constrained by the type of data that they can process, LLMs excel at representing large amounts of heterogeneous data in a simplified, uniform way. By reducing the complexity of data, these models could therefore act as assistants to existing AI models.
One source with knowledge of the fundraising says that Bioptimus is creating an LLM that will be used as a way to simplify complex biomedical data for existing AI technologies, such as Owkin’s predictive AI model. This will help these models make sense of wide-ranging data, improving their efficiency and accuracy.
While Bioptimus will focus at first on enhancing AI models in the pharmaceutical field, its LLM could eventually be applied to many other industries, such as foodtech or cosmetics, according to the same source.
Troves of patient data
In a field where access to information is heavily regulated, Owkin is sitting on volumes of high-quality patient data. The company’s predictive AI models are based on data that it sources from partnerships with academic medical centres.
This is a significant competitive advantage for Bioptimus, which will be able to use Owkin’s datasets to train its model.
HQ'd in France and in the US, Owkin has raised more than $300m to date since it was founded in 2016 from investors including partner companies Sanofi and Bristol-Myers Squibb, as well as French VCs Frst, Cathay Innovation and family office Otium Capital.