Thèse Vers des Modèles Robutes en Biotechnologie H/F doctorat.gouv.fr
- Paris - 75
- CDD
- Service public d'état
Détail du poste
Établissement : Université Paris-Saclay GS Sciences de l'ingénierie et des systèmes École doctorale : Sciences Des Systèmes Laboratoire de recherche : LGPM - Laboratoire de Génie des Procédés et Matériaux Direction de la thèse : Victor POZZOBON ORCID 0000000315300834 Début de la thèse : 2027-10-01 Date limite de candidature : 2026-12-31T23:59:59 Ce projet doctoral a pour but d'apporter au monde de la biotechnologie des méthodes de modélisation capables de tenir compte de l'incertitude inhérente au vivant. En effet, lors de la réalisation d'expériences l'incertitude (qu'elle soit de mesure, ou d'origine biologique) rend l'interprétation plus complexe et affaibli la confiance dans les résultats. Pour y pallier, les expériences sont répliquées et validées avec des outils statistiques. Cependant, lors des étapes de modélisation, de tels outils adaptés à la biotechnologie ne sont pas disponibles facilement, voire n'existent pas. Par conséquent, la propagation des incertitudes peut conduire à des divergences des modèles rendant leur emploi pour le design et le contrôle des bioprocédés sujet à caution. Il est donc urgent de les développer afin d'augmenter la confiance dans les modèles et surtout la robustesse de leurs prédictions. We believe there is an urgent need for tools that help researchers and engineers better manage uncertainty (defined as the width of the confidence interval surrounding quantitative elements) throughout the upscaling of biotechnological processes. This opinion is based on our own experiences, interactions with peers at conferences, informal discussions, project meetings, and bibliometric analyses. The absence of such tools has three consequences:
- Scale-up procedures are cautious (gradual increase in process size, with a design of experiments at almost every step), and therefore time-consuming and costly.
- Model predictions are qualitative or require an extensive (and costly) database to be reliable.
- The reproducibility of experiments is not guaranteed.
This urgent need for new tools is reinforced by the scarcity of literature on the subject. For example, the Web of Science database was searched using the keywords uncertainty and scale-up for the field Biotechnology & Applied microbiology. Only 31 documents mention these two keywords, 22 of which are research articles, and only half are technical articles (the others being economic analyses or irrelevant documents). Furthermore, mentioning uncertainty does not mean that techniques need to be developed to deal with it.
For example, the following figure (see attached pdf) shows the behavior of a system that does not take uncertainty into account in its design (left) and a model that does (right). It clearly shows that taking uncertainty into account is crucial during the design stages. In this context, this doctoral project aims to bring uncertainty management methods within a unified conceptual framework to the world of industrial biotechnology. It will develop digital tools that facilitate upscaling processes in industrial biotechnology by providing better models while minimizing the need for new small-scale experiments (saving time, reducing costs, etc.). The thesis will focus on three areas: the implementation of methods to take uncertainty into account in modeling, the development of methods for recommending experiments to reduce these uncertainties, and the application of tools to a specific (simple) case. The bibliography will support the investigations from start to finish and facilitate the publication of our results.
Literature review
The objective of this study is to update students on the following areas:
- modeling in biotechnology,
- methods for managing uncertainty in experimental approaches,
- methods for managing uncertainty in other fields,
- an area of application to illustrate the methods (chosen by the candidate, with microalgae, yeast, and bacteria being the team's core areas of expertise).
Managing uncertainty in modeling
We will develop three types of concepts: the classification of uncertainties (critical-leading to divergence-moderate-amplified by the system, but controllable-acceptable-mitigated by the system-etc.), the quantification of uncertainties (local sensitivities, global sensitivities, confidence intervals, etc. (Sobol, 2001)), and their consideration in model construction methods. These are ambitious challenges. To meet them, we will draw on other scientific communities that have begun to address the issue of accounting for this uncertainty in their models. For example, at the level of the objective function (or cost or loss function), which must be minimized when estimating model parameters based on a set of experimental data (Croke, 2007). Or at the level of quantifying the extent of confidence intervals for estimated parameters via Fisher's information matrix (Transtrum & Qiu, 2012).
However, none of these approaches are used in biotechnology. The objective will be to initially build on these solid foundations to propose methods adapted to our field of study. Of course, we will not limit ourselves to these three methods; new ones will be proposed. Finally, an effort will be made to unify them before passing them on to the community.
Recommendation of experiments
Once the above concepts have been established, we will focus on how to apply them. The idea will be, for example, to see how they can be used together to recommend new experiments that will help reduce uncertainty. Ideally, we will be able to develop optimal methods that recommend the most interesting experiments in succession (Jeong et al., 2018).
To make this concrete, we will apply our tools to in silico problems. For example, with a hidden model and a learning system trying to model it. We will also compare our tools with existing ones, such as design of experiments methods.
Application example
A model microorganism will be chosen to illustrate the capabilities of the methods developed. We will set up simple experiments to feed biological behavior models. In this way, we will challenge our new tools with this system and demonstrate the applicability of our conceptual proposals in real-world conditions. Specifically, we will set up a system using bacteria, microalgae, or yeast, depending on the preferences of the doctoral student leading the project.
During the thesis
Regular reports, in the form of written reports and oral presentations, will be expected from the doctoral student. This will allow them to accumulate material that has already been written, thereby facilitating the writing of their manuscript. At least one publication in an international peer-reviewed journal will be a prerequisite for the thesis defense.
Le profil recherché
Publiée le 20/11/2026 - Réf : 578198c8a736e657d36513cc8b263d9b