Hello Julien, nice to meet you! Can you start by introducing yourself and Scienta?
My name is Julien Duquesne, I am an engineer and researcher working on Artificial Intelligence. I am 28 years old and I am the co-founder and CTO of Scienta.
“Scienta is a company developing a foundation model of the immune system to predict patients’ responses to newly developed drugs before their efficacy and safety are assessed in a clinical trial. ”
Our long-term goal is to simulate in silico a patients’ response to any treatment and increase success rate of drug development processes in immunology field.
Scienta is a TechBio and not a Biotech, isn’t it?
Exactly. We don’t develop our own treatments, which is what biotechs do. At Scienta, we are developing a technology, a solution that we will sell to Biotech and Pharma companies to help them develop their treatments. We are technology-providers. This is an important distinction: we are experts in technology at the service of therapeutics.
How did you end up joining or creating Scienta with Camille and Vincent Bouget, respectively CEO and CSO of the company?
The company was born from a project Camille was leading. While working in Pharma, she realised that in the field of immune-mediated inflammatory diseases, we don’t always know why patients respond or not to the treatments they are given. There can be high rates of treatment failure and disease flares.
We need new models to better understand this phenomenon and tackle it, to increase the effectiveness of existing drugs but also to design more effective ones. The idea of precision medicine, personalized medicine, is what gave birth to Scienta.
What’s your story as CTO? What ups and downs did you experience while leading the technical section of the company?
Scienta’s story started 5 years ago, and we’ve come a long way since then. Initially, I am not a healthcare professional or biologist – I come from AI and informatics, which is precisely why I was attracted to the project. I think health is the best field in which to do AI for multiple reasons: first because there is a concrete impact – to improve patients’ lives, but also because this is a field where we are pushing the limits of human knowledge. AI is not only a vector for productivity but also for discovery, which is extremely stimulating.
Regarding the ups and downs: building AI models is very research-oriented, almost like a PhD program. We test approaches which sometimes work, sometimes not. As co-founders, we’ve had a lot of pivots over the years – always towards precision medicine in immunology, but not necessarily with the same datasets or value proposal. We started with clinical data with physicians, and now we are focusing on the translational side with researchers, notably omics data. It’s a non-linear path, but this is how you learn how to find the right direction.
It is said that all good things come to an end, but is that true for a company or a project? What kind of entrepreneur are you: will let the “bird fly the nest” and start building something new? Or will you stay and capitalize on your work with Scienta?
I don’t really think about it in this way. My objective is to have the maximum impact on society, patients, and science. So as long as I am in environment that allows me to do that – with brilliant people around me, resources, challenges – I am really happy to stay. For now and since the beginning, Scienta is an environment which allows us to exactly do that. We’ve been successful, especially on the scientific side.
It is not about letting go of a project or not – it’s about asking if we’re going in the right direction. For now, the answer is yes.
Since we aren’t all AI experts, can explain what a foundation model is in the context of drug discovery in immunology research?
This is a new paradigm shift that emerged around 10 years ago. Classically, for each project, data set, or task, we run a new model – whether that’s a logistic regression, random forest, neural network – which would be predictive on a specific task. The main issue with these approaches is that they are limited by the available data. In biology we often have cohorts with a small number of patients, but with many variables. Statistically, it’s not easy to extract some signal!
The hypothesis behind a foundation model is to say: if we show it very large amounts of existing data – not necessarily labelled – can the model learn patterns, representations which then help it perform a specific task? This is exactly what LLMs do with the language: we don’t give them a specific task, they learn the relationship between the words and then transfer this learned knowledge to new task.
In biology, if a model has already understood that what a cell, a tissue, or a patient are, there will be less to learn again when we give it a precise task. This is a very promising approach.
You have an impressively multidisciplinary team, including pharmacists, engineers, biologists… What has been place of PhDs place in the project?
Multidisciplinarity is crucial here because we are at the crossroads of multiple sciences. Having health and biological data experts has been absolutely central – people capable of truly understanding the data we are handling and having a critical view on our results and their interpretation. We put a huge importance on our model’s interpretability, to ensure our results make sense biologically. This is not just a model which says yes or no without us knowing what is happening inside it.
This requires the expertise of immunology and biology PhDs, working hand in hand with engineers and AI researchers. We quickly realized we couldn’t have it any other way!
What are the strengths of your foundation model?
A major strength of foundation models, and ours in particular, is that it works with little data. Even from limited preclinical data, from a few mice for example, the model can produce a first estimation of patients’ responses and with a signal about probabilities of success.
Then, we have scalability. Classical methods like QSP model biology with available knowledge from the literature, but these must be re-created for each new drug and disease – which is very long and expensive process. Our model has already learned some patterns that it can directly apply to a new treatment under development.
Finally – and this is a strength that is sometimes underestimated – by training our model directly on biological data, and not on language like LLMs, we avoid this language bottleneck and we get much reliable and accurate biological representations.
“Being truly capable of working with scarce data is a huge strength of foundation models in comparison with classical statistical methods.
What are the weaknesses of the model? Do you only overlap histology and transcriptomics data? How is it evolving?
We don’t yet integrate all relevant modalities in immunology, though of course that’s the end goal. We integrated proteomics data since the beginning of the year for instance. The aim is to get a complete picture of the immune system spectrum at different scales, because linking all scales between them is what will give us an accurate panorama.
In addition, a limit common to many AI solutions is computing resources. These models are highly needy in GPU, as much for their training as for their use, and they require a certain expertise to be both created and used in practice. We cannot just put them into the hands of biologists who know nothing about AI, unlike LLM models where you can just ask a question to get an answer.
Right now, we are building a true biological model – you start from zero, you take the data and you are trying to replace and connect the pieces of the puzzle together and have a global vision. It is more demanding, but that’s exactly what makes it more rigorous.
What is the added value of foundation models? They seem to be really good at identifying targets, but can we imagine making new scientific discoveries?
I think we are pursuing both these things at the same time – both are in fact well connected.
The main advantage of the foundation model is that it is very versatile in terms of tasks. We can focus on highly predictive task – predict the effect of treatment – but, thanks to the explainability of our tools, we don’t stop there. We want to understand what about a patient’s immune system makes them a better responder to a specific drug and potentially uncover new biological mechanisms.
We had a very concrete illustration with our publication in The Lancet Rheumatology last year. We went from a predictive model for Sjögren’s syndrome diagnosis, and by diving into the data – in collaboration with rheumatologists and immunologists– we brought to light new patterns and biological mechanisms which bring us new understanding of the disease. This process – build, interpret and discover – is the strength of what we do.
Finally, what is the best advice you would give to an undergraduate student in Biology, a Master student and a PhD student, respectively?
I haven’t done any studies in bio, so it is a bit complicated for me to put myself in their shoes, but here is what I think.
For an undergraduate student in biology: the most important will be to go and understand things by yourself. It is even more important at the age of LLMs – to have true critical thinking, asking the right questions, digging into them, and to not take for granted any things we have read. You have to be able to do the reasoning again, not just consume knowledge. I do a parallel with maths preparation classes: this is what we learn – understand, don’t just memorize.
For a Master student : this is the moment to start having a critical view of the state of the art, identify questions that are truly important for research, which can give rise to a PhD, and face them as soon as possible.
For a PhD student : Today, being able to use AI is a prerequisite, do not hesitate. You should consider biology as a multidisciplinary field where we will need more bioinformatics, AI, and data sciences. Pure bio PhD are going to be more and more rare. This is going to be multidisciplinary effort, which is a great opportunity!
A successful first edition of European Doctoral Day
This May, the first European Doctoral Day brought together a series of events dedicated to shine a light on PhDs and their role in innovation and society.
As part of this European initiative, Réseau Biotechno was delighted to help organise one of these events, hosted by Sorbonne Université and broadcast live online, alongside partners who share the same mission.
Over 100 participants joined us and questioned the role of PhDs in facing the challenges of our time, through interactive workshops – theatre, quizzes, debates and even a trial!
In the words (translated to english) of Dr. Sylvie Pommier, who coordinates the doctorate at the DGESIP-DGRI, “It was original, dynamic and fun, and carried real core messages”.
The take home message?
During their doctoral studies, PhD develop critical thinking skills, scientific methodology, deal with uncertainty, and produce new knowledge. These are exactly the assets needed by the socio-economic world to innovate and adapt to new challenges!
We would like to thank all our partners, everyone involved in organising this very first edition, and all the participants who brought this day to life.
The European Doctoral Day initiative will continue across Europe in 2027.
As a partner, we are delighted to support BioFIT Event and MedFIT Event!
Join events in Marseille on December 10th and 11th, 2026 for the joint edition of Europe’s leading partnering conventions dedicated to innovation in the Life Sciences and Healthtech sectors!
On the agenda:
- Opportunities for strategic collaborations with over 1,300 international delegates
- An inspiring conference programme with 100 international speakers
- A dynamic exhibition area with 160 exhibitors
- Presentations of innovative projects
Thanks to our Réseau Biotechno partnership, take advantage of an exclusive 20% discount for our members! For more information, and to receive your discount code before registering, please get in touch with your contact in our organisation.
Discounts are always applied to the ‘regular’ fee except during the ‘late registration’ period. Discounts cannot be combined with other offers.
Nuits des étoiles (star nights)
If the nights are clear on 7-9th of August, take a look at the night sky! Over 500 free astronomy events will take place across France celebrate this special occasion to see the stars.
Sur les épaules des géants
This science festival aims to bring together scientists and the general public, with workshops, conferences, and fun meeting events! Event held in French.
Fête de la Science
The French Ministry for Education, Research and Space hosts “Fête de la Science”, with this year’s theme being “Saveurs savantes” loosely translated as “Tastes and knowledge”. Check out hundreds of free scientific events, mostly held in French.
