Helical, one of Luxembourg’s most promising AI healthtechs, has just raised a €2.2m Seed to build its open-source platform of bio foundation models, aiming to accelerate pharma research and development. Maxime Allard, co-founder and CTO of Helical tells us more about the platform’s aims, the struggles of getting set up in Luxembourg and why comparisons to ChatGPT are misguided.
Co-founded in 2023 by Luxembourgers Rick Schneider (CEO), Mathieu Klop (CSO) and Maxime Allard (CTO), Helical has already captured the attention of local and international investors, raising €2.2m in seed from Frst, a French VC, BoxGroup and notable Business Angels from world-leading AI companies. Among them are the co-founders of Cohere ($5bn AI company), Aidan Gomez and Ivan Zhang, and the co-founder and CEO of Huggingface ($4.5bn AI company), Clément Delangue.
While last year the startup was still learning the ropes of entrepreneurship, this year it seems to have found its footing and is ready to open its platform to the community of biologists, bioinformaticians and researchers who are keen to leverage it for their research goals. Maxime Allard, co-founder and CTO of Helical tells us more.
How would you explain Helical to someone with no background in biology or AI?
Helical is a kind of intermediary platform which allows biologists and bioinformaticians to have access to the best tools from AI and computational biology at their fingertips. They don’t need to think about how to use AI or how to code, they can simply use our platform and with a few clicks, they will have access to the best bio foundation models in the world which they can use on their data.
Helical is the first open-source platform for data scientists at pharma companies to experiment with, benchmark and build on single-cell & genomics Bio Foundation Models. We are the extended arm between the open-source community and bridge the gap between open-source models and applications.
Comparisons between Helical and ChatGPT have already been made. How accurate are they?
I think it’s an easy comparison to make but I don’t immediately agree with it. The inputs into Helical will not be text like in ChatGPT but rather they will be DNA and RNA samples which can be coded in different formats. In the beginning, the open-source package will be in Python, so these user uploads will be done with Python-APIs. Our package would allow users to make statistical calculations with simple API calls and extract results very simply. So this is very different from ChatGPT because it’s not as reactive.
In a second phase, we would be open to considering adding an interface where biologists, who are less familiar with coding, could actually upload files and get results that would create a feedback loop similar to the one in ChatGPT.
“We want to make our platform so easy to use that researchers, biologists and technicians can use it in their daily work.”
Maxime Allard, co-founder and CTO of Helical.
What is the added benefit of this platform?
We are seeing a lot of new applications emerge with these AI models (also called bio foundation models in this context) because they can better recognize patterns in DNA, which has three billion nucleotides. Before those models, analysing and identifying patterns in these complex datasets was close to impossible and often a trial-and-error process. We aim to change this.
Today, with the help of these models, one can look at a few million nucleotides simultaneously and therefore extract patterns and results a lot faster. We’re moving from trial-and-error research that was done in the lab (in vitro) to being able to do the same with a machine (in silico). This is a game changer because it allows us to obtain results faster, cheaper, and be more precise.
What’s the first step towards opening Helical up to potential users?
To get started, we want to give users access to a list and user manual of the best bio foundation models on the market. If they have already created their own foundation model, our platform will allow them to benchmark theirs. There will be many different tasks for which they will want to test their models: biomarker discovery and target prediction being just two of the more popular ones.
We help companies leverage existing open-source foundation models. For the moment, the idea is not to create these models ourselves, as this is a very capital-intensive process.
What is Helical’s main commercial use case?
I think the main value lies in the improved application of your models that our platform will provide. We want to make our platform so easy to use that researchers, biologists and technicians can use it in their daily work. Integrating our platform into their workflows and listening to our clients’ feedback will be one of our key goals.
What has been your experience of launching Helical in Luxembourg?
Creating a startup in Luxembourg has been very, very difficult at times. It took us four months to get a proper bank account to get started. Another thing which is hard and which we are currently working on is creating a stock options plan to make the Helical more attractive to future employees. In Paris, this is something that companies have already sorted out really well, but I know that Luxembourg is also working on a better solution.
You have already partnered with LuxProvide’s startup programme, NVIDIA’s Inception programme, and Microsoft’s Founder Hub. What can you tell me about the details of these partnerships?
The nice thing about LuxProvide is their very secure, business-ready supercomputer that allows us to get started on our co-innovation partner projects, get useable results and do this at a very large scale. The NVIDIA Inception Inception programme is really interesting because we can start getting hardware for GPUs, for example, at reduced price prices. We have access to the latest platforms they have, including those on genomics. So this allows us really well to go deeper into that NVIDIA ecosystem. Microsoft Azure is really interesting because you actually have a lot of the things you need to get your startup up and running including LinkedIn subscriptions, Github enterprise and OpenAI credits among many other things.
What do you hope to have achieved by the end of the year?
By the end of the year, we want to be the de facto standard for people to compare their benchmark models against. We also want to be present both in academia and in industry and build our community around our users. We also hope to have 5 to 10 partnerships with bigger companies and develop new use cases to gain as much traction as soon as possible. Our funding round will also allow us to grow the company to 10 people by the end of the year.
