Déborah Boyenval

About Me

I’m currently a postdoctoral fellow at the University of Helsinki in the Systems Biology of Drug Resistance in Cancer group (Hautaniemi Lab). Since 2025, I have been developing mathematical models using data from the DECIDER cohort of patients with high-grade serous carcinoma. My postdoctoral research comprises two projects:

• Generative modeling of synthetic multi-omic cohorts: I develop and evaluate generative models to produce synthetic multi-omics data (typically genomic, transcriptomic and methylomic data) for sharing with an industry collaborator, Orion Pharma, with applications in precision medicine and drug discovery. Our paper presents a biologically motivated framework for evaluating such synthetic data. It was accepted for oral presentation at the GenAI4Health workshop at NeurIPS 2026 and as a Findings paper at ML4H 2026.
• Modeling tumor evolution: I simulate virtual tumors using branching processes to investigate the mechanisms underlying treatment resistance.

Previously, I was a postdoctoral researcher at the LaBRI in Bordeaux, where I worked within the Synthesis team. Coming from a Ph.D. in theoretical computer science, I wanted to broaden my horizons by exploring how modern biological data, including high-throughput sequencing and multimodal datasets, could constrain the parameterization of discrete mathematical models of biological systems. More specifically, I explored how to personalize logical models using mouse multi-omics data (including RNA-seq, CITE-seq and ATAC-seq) to identify the regulatory networks underlying hematopoietic phenotypes and investigate how aging reshapes their structure.

During my Ph.D., I worked in the field of formal modeling of biological systems, also called symbolic systems biology by the GT-Bioss, or “symbolic AI”. Monthly seminars are organized so feel free to take a look if you’re new to the field! My Ph.D. was in theoretical computer science (e.g. formal language and automata theory, computability theory, complexity theory and algorithms), applied to biology. The guiding idea of my thesis was to verify the correctness of a biological regulatory network just as one would verify a computer program! My Ph.D. thesis focuses on the discrete modeling of intrinsic cell cycle checkpoints and its formal verification. It was supervised by Gilles Bernot, Jean-Paul Comet and Franck Delaunay. If you’re curious, you can find more details about my thesis below.

Research Interests

• Generative methods: variational autoencoders and transformers for modeling multiomics data.
• Bio-LLM: I am increasingly exploring the use of large language models to investigate and organize biomedical knowledge, as well as support evidence-based reasoning. My goal is to connect mechanisms described in the literature with patient multi-omics data to systematically generate and evaluate testable biological hypotheses, particularly about treatment resistance in ovarian cancer.
• Mathematical modeling: discrete, continuous or stochastic approaches depending on the biological question, with expertise in discrete models, formal methods, PK-PD models based on ordinary differential equations, and branching processes.
• Multi-omics data analysis: integration and interpretation of multiple molecular data types.
• Cancer biology: modeling so investigation mechanisms for tumor evolution and resistance phenotypes, drug action mechanisms.
• Causal modeling: a growing interest in causal inference and do-calculus, analyses beyong correlations.

More about my Ph.D. thesis

My Ph.D. thesis was initially motivated by a need to model the coupling between the circadian cycle, the cell cycle and metabolism, with the prospect of (better) understanding the circadian chrono-toxicity and efficacy of anti-cancer drugs. During my first year as Ph.D. student, I repositioned the problem on the question of the dynamic specifications of such a coupling. What are the dynamic characteristics of the coupling itself, apart from the union of the dynamic characteristics of the three modules?

I propose a focus on the notion of biological checkpoint, which is described very intuitively as a set of steps that must be completed before starting another. This discrete notion applies reasonably well in the case of cell cycle regulation, since it is described dynamically by a set of switch-like molecular events of activation and inhibition of the cell cycle core regulators. A checkpoint is thought here as a constraint on the order of some blocks of cell cycle events named phases. Here we refer to the phases :

• S: DNA replication,
• M: Fair distribution and division of DNA
• G1 (prior S) and G2 (prior M): gap phases for energy and biomass resources.

DNA replication and division are highly regulated processes. Errors in DNA copying or distribution in descendant cells must be limited for the survival of the species. A widely cited example cell fusion experiments show that certain phases cannot occur simultaneously and that there is an order of phases to preserve DNA integrity (great scheme here). A G2 cell, ready to enter M phase, is merged with an S phase cell. The G2 cell halts its progression and does not enter M phase. In addition, the G2 cell does not start a new DNA synthesis phase. This G2 cell appears to be temporally isolated from the S and M phase. This remarkable experiment shows that :

• The G2 –> S transition seems to be possible only when a new cell cycle begins, i.e. only after the M phase, so the transition S/G2 is irreversible (unidirectionality).
• The G2 –> M transition is only possible when the S phase has ended, this is an example of phase order constraint.

If we reason in terms of transition system, two phases x and y (in blue) are temporally isolated if it is not possible to reach y from x by any path without passing through a transition state that certifies that phase x has ended (in green). Furthermore, there is no way to reach x from y without passing through a state of y that certifies that y is a completed phase (in green).

Discrete formalisms for modeling biological regulatory networks, such as boolean/multivaluated networks or Thomas’ formalism, are well suited to the study of temporal phase separation, described by these forbidden paths between phases. Thomas’ formalism has been coupled with formal methods for selecting discrete models, such as model-checking or Hoare’s logic. The idea was simple: only those models are selected which satisfy the interaction graph given as input, as well as dynamic specifications on observed biological traces (Hoare logic) or broader dynamic specifications such as CTL. The interaction graph of mammalian cell cycle progression used for the proof-of-concept was entirely renovated from Jonathan Behaegel’s seminal model, itself an update and discretization of Tyson and Novak’s model. I have abstracted each phase of the cell cycle by a set of activation or inhibition events of the cell cycle regulators from the rich literature available on the subject. Based on these phases specification that have been formally proven correct using Hoare logic, it was then possible to reason about the notion of temporal separation with model-checking of CTL formulas. Thanks to all these materials, my Ph.D. team and I were able to model the checkpoints G1/S, S/G2, G2/M and the metaphase/anaphase transition checkpoint, thus proving that a checkpoint is a discrete, non-continuous, concept.

Would you like to find out more? Go to the Publications tab below.

A checkpoint in the broadest sense constrains any type of event. I took the example of the cell cycle in my thesis since the problem focuses on a single physiological system which turns out to be the most documented example of checkpoint-regulated system in biology. Complicating the issue, metabolism is intrinsically involved in the progression of the cell cycle. The circadian system coordinates physiological processes over 24 hours. Apart from the chronometric aspect, my thesis poses an open question: does the circadian system act as a checkpoint (i.e. a temporal separator) between the phases of the cell cycle (G1, S, G2, M) and metabolism (oxydative, reductive)?

News

• [October 2026] Our paper has been accepted for an oral presentation at this year’s GenAI4Health @ NeurIPS 2026 workshop in Sydney! It has also been accepted in the Findings track of the ML4H 2026 Symposium.

Publications

Services

Teaching experience

• 2025–2026 — University of Helsinki: Contributed to the Introduction to Systems Biology course (Spring 2025). Teaching relational databases as part of the Clinical Data Mining course (Fall 2026). Both courses are coordinated by Sampsa Hautaniemi.
• 2018–2021 — Université Côte d’Azur: Teaching Assistant at BSc and MSc levels. Courses covered web development, systems and network administration, database theory and SQL, imperative programming and Python, and algorithms for biology.

Peer review

• PLOS Computational Biology
• ML4H

Science outreach

• [September 2022] A conference for Woman In Machine Learning (WiML) on the principle of model checking and a practical application in biology. You can find the slides here.

Graduate school

• [2020 - 2022] Treasurer of the DS4H graduate school student association
• [2021 - 2022] Elected student member of the DS4H graduate school scientific and pedagogical council.

Talks and posters

• [Talk, December 2026] Synthesis of Ovarian Cancer Multi-Omics Datasets: A Biology-Informed Evaluation Framework at the GenAI4Health Workshop at NeurIPS 2026 (Sydney).
• [Poster, December 2026] Synthesis of Ovarian Cancer Multi-Omics Datasets: A Biology-Informed Evaluation Framework at the ML4H 2026 Symposium (Sydney).
• [Poster, 2026] Generating Synthetic Data from Ovarian Cancer Patients at RPU Science Day 2026 (Helsinki).
• [Poster, 2025] Learning Resistance: Synthetic Data for Predictive Modeling in HGSC at the DECIDER Annual Meeting (Helsinki).
• [Talk, December 2023] Formal modeling of the discrete dynamics of intrinsic cell cycle checkpoints: a proof of concept at GT-Bioss (online).
• [Ph.D. Defense, December 2022] Formal modeling of biological cyclic behavior with checkpoints: the cell cycle regulation at I3S laboratory.
  (Slides).
• [Talk, September 2022] Talk on the principles of model checking and a practical application in biology for Women in Machine Learning (WiML) (Jussieu, Paris).
  (Slides).
• [Talk, March 2022] A discrete modeling study devoted to the formalization and verification of mammalian cell cycle checkpoints at Lifeware Public Seminar (Inria Saclay, France).
  (Slides).
• [Talk, November 2021] Mammalian cell cycle, formalising phases at GT-Bioss Annual Days (Lyon, France).
  (Slides).
• [Poster, July 2021] Logical and incremental formalization of cell cycle checkpoints at JOBIM 2021 (held online due to COVID-19).
  (Poster).
• [Talk, June 2021] Logical modeling in biology through the case study of the cell cycle and its checkpoints at a public seminar of the Institute of Biology of Valrose.
  (Slides).
• [Talk, November 2020] Study of cell cycle checkpoints, specification and verification at GT-Bioss mensual seminars.
  (Slides).
• [Talk, September 2020] What is a cell cycle checkpoint, The TotemBioNet answer at CMSB 2020.
  (Slides).
• [Talk, December 2019] From phase characterization toward observable properties verification, at NeuroMod Institute for a Public Ph.D. Seminar (Sophia Antipolis, France).
  (Slides).
• [Poster, November 2019] Introduction of Priorities in Biological Regulatory Networks using Thomas’ Formalism at Modelife Annual Meeting (Fréjus, France).
  (Poster).
• [Poster, June 2019] A discrete cell cycle model: from phase characterization toward observable properties verification, at a CNRS summer school about formal modeling of biological regulatory networks (Porquerolles, France).
  (Poster).

Links

• Gilles Bernot (my former Ph.D. supervisor) and the youtube channel “Qu’est-ce que tu GEEKes?” made a superb video about Gilles’ job as a university professor. His witty stories and remarks accompanied and nourished me throughout my thesis. Go this way.
• I’d like to share the websites of people I’ve met and who have inspired me over the years in my research : Julien Martinelli, Athénaïs Vaginay and Stéphanie Chevalier. I also recommend the website of Chloé-Agathe Azencott, which is full of resources for machine learning newbies.

Powered by Jekyll and Minimal Light theme.