Research engineer (CNRS permanent position)

A CNRS research engineer position in software development and AI is open in our team. Interested candidates should rapidly contact us (applications  on the CNRS site end on July 10, 2024). 

Poste ingénieur recherche

Junior Professorship

Tenure track

PhD Scholarships

PhD offers



Internship :Hierarchical modeling of melanoma cell lines

Start and duration: as soon as possible for at least 4 months.

Location: Computational Systems Biology team, LPHI, University of Montpellier.

Candidates: M1 or M2 students / and or engineering students with excellent background in computer science or mathematics and a taste for interdisciplinary research such as application on biology. The intern will receive a salary (gratification) at the standard rate used by French public research institutions.

Skills required : good dev skills in Python (core Python expertise, machine learning, debugging, version control), algorithmic problem solving, good communication skills in an international project.

Context: Thanks to new technologies and shared databases, access to large quantities of biological data has become widely available. Today, we have access to information gathering several dimensions of data (multi-omics, imaging, time-series, single cell, different individuals). Faced with this boom, artificial intelligence and machine learning tools have been developed considerably over the last few years, but remain for the most part black-box tools that do not allow us to account for how data is taken into account in predictions. Alternatives are provided by mechanistic models that represent white-boxes approaches, or by hybrid settings combining back and white boxes.

Two hurdles prevent substantial scaling up of mechanistic models: the lack of standardized automatic generation of models and the absence of effective methods for determining the parameters of these models. Within an international CEFIPRA project, with our indian partners from NCBS Bangalore (team of U.S. Bhalla) we have recently developed a computational framework for hierarchical modeling and parameter optimization of mechanistic models. In this framework mechanistic models and different types of experimental data are represented at different levels of abstraction in a standardized way. In hierarchical modeling, we divide a complex problem concerning large models into simpler subproblems that are solved recursively. In biology, this is made possible by the modularity of biochemical pathways.

Goal: We aim to develop automatic tools for passing from one level of abstraction to another one and for modifying a given a model by making it consistent with other modeling hypotheses. Three levels of abstraction are considered here: weighted interaction graphs with modular response analysis capacity, Hill-tau event based abstractions, and ordinary differential equations models. The relations between different levels of abstractions (model reduction, or model complexification) will be addressed algorithmically. Our existing hierarchical optimization methods for parameter identification will be extended to encompass the case of hierarchies combining several levels of abstraction.

As a case study, we will apply this framework for the analysis of drug sensitivity of melanoma cell lines, using genomic, transcriptomic and phosphoproteomic data. We expect to specialize the models to different cell lines, and obtain disease-specific models. Furthermore, our strategy is in line with FAIR principles in developing complex models that can be easily reused.


Radulescu,O., Gorban,A.N., Zinovyev,A. and Noel,V. (2012) Reduction of dynamical biochemical reaction networks in computational biology. Frontiers in Bioinformatics and Computational Biology 3: 131.

Viswan, N. A., HarshaRani, G. V., Stefan, M. I., and Bhalla, U. S. (2018) FindSim: a framework for integrating neuronal data and signaling models. Frontiers in neuroinformatics, 12, 38.

Viswan, N. A. and Bhalla, U. S. (2023) Understanding molecular signaling cascades in neural disease using multi-resolution models.

Current Opinion in Neurobiology, 83,

Viswan, N. A., Tribut, A., Radulescu,O., and Bhalla, U. S. (2024) Hierarchical Optimization of Biochemical Networks (preprint).

Nyman, E., Stein, R. R., Jing, X., Wang, W., Marks, B., Zervantonakis, I. K., Korkut, A., Gauthier, N. P., and Sander, C. (2020) Perturbation biology links temporal protein changes to drug responses in a melanoma cell line. PLoS computational biology, 16(7), e1007909.

Keywords: machine learning and hybrid AI, systems biology, mechanistic modeling, oncology data

To apply: send email + CV + transcripts to Ovidiu Radulescu

Older offers, some are still valid

  • CEFIPRA project: Automated generation and analysis of signaling networks models for quantitative biology and precision medicine

Summary of the project

Subcellular signaling controls all aspects of cellular function in growth, differentiation, plasticity, health and disease. Models of these signaling networks have applications in fundamental biology and medicine. However, it is difficult to make and validate such models, due to the complexity of the signaling systems and the heterogeneity of the data. Most such models are currently manually developed. We propose algorithms and tools to generate models with different levels of abstraction, going from molecular interaction maps to detailed, quantitative chemical reactions networks. Our approach will be to a) acquire data from databases and the literature, and b) to develop models at the successive levels of interaction maps, abstract reduced pathway models, and detailed mass-action models. Our specific innovation will be to automate the hitherto challenging steps of going from interaction maps to reduced models and thence to mass-action models. We propose to do this using automated exploration of model topology space, starting from block-diagram databases, literature mining and phosphoproteomic datasets. We will utilize extremely efficient model simulation and fitting to select models. We will utilize formal model reduction and analysis methods that we have recently developed. Our deliverables will be 1. A pipeline for generating well fitted and parameterized models at multiple levels of abstraction, from diverse data sources. 2. Tools and theory by which more abstract models can be used to constrain more detailed/mechanistic ones and 3. Applying these approaches we will develop prototype models of two specific disease systems, namely Autism Syndrome Disorders, and Melanoma, across the range of abstraction levels. Our project will open up a path for progress into trustworthy, explainable AI, that combines mechanistic models and big data for transparent decision making in medicine.

For this project we will hire a post-doc and several interns. 

1. The post-doc call 

Computational Systems Biology LPHI (University of Montpellier and CNRS) invites applications for a postdoctoral position founded by CEFIPRA (Indo French Center for the Promotion of Advanced Research) in collaboration with Upinder S. Bhalla’s lab in NCBS Bangalore. We seek a candidate with expertise in quantitative biology, simulation, dynamical systems and machine learning. The contract is for one year with the potential of further extension. The successful candidate will work with the French partner. He/she will develop novel hierarchical methods for learning dynamical models of signaling using biomedical data.  

2. The internship call

We are looking for interns with background in computer science for developing tools for automated construction, manipulation, visualization, simulation and machine learning of biological signaling models. These models are of different types (directed  graphs,  neural networks, chemical reaction networks) and can be simulated using different formalisms (ordinary differential equations, event  driven systems, stochastic models). The models have parameters that are learned using time-series data. The tools will be developed in Python.       


  • Computer algebra project in automated model reduction (one internship)

We are looking for an intern with expertise in computer algebra to implement model reduction methods for biochemical networks. These methods were developed in the ANR project SYMBIONT and are based on tropical geometry, singular perturbations and parametric rank calculations. Preliminary versions of the model reduction tools were developed in Python and Maple. The intern will learn about innovative methods in scientific computing applied to biology and medecine.


To apply: send CV + motivation letter + transcripts to

  • NEW! MALMO (several internships)

MALMO is financed by the MIC (Mathématiques et Informatique – cancer) program of Itmo Cancer from November 1, 2020 until October 31, 2023. We hire several interns on various topics related to mathematical oncology and artificial intelligence. The internship duration is between 4 and 6 months (extensions are possible) to start as soon as possible and not later than March 2023. The internships are also tightly related to international collaborations with the team of Upinder Bhalla in NCBS Bangalore and Holger Fröhlich in Franhoffer SCAI Bonn

Biological background

Metabolic rewiring is a recognized hallmark of cancer cells. The metabolic reprogramming of transformed cells is required to sustain proliferation and biomass production, but it also influences many other biological processes, including cell signalling or the epigenetic control of gene expression. Many metabolic alterations of cancer cells influence their sensitivity to chemo- and targeted therapies but the underlying mechanisms are not fully understood. Directly relevant to our project, it is well recognized that this metabolic reprogramming is influenced by variable microenvironmental conditions, including nutrient and oxygen availability. Our objective is to further understand cancer heterogeneity in melanoma from a metabolic standpoint and to evaluate how gradients of nutrients and oxygen influence melanoma cell fate and drug resistance.

Mathematical modelling background

Cancer is a complex disease involving multiple genetic and epigenetic changes that continuously evolve during disease progression. In order to survive and proliferate, cancer cell populations use adaptive evolution strategies based on heterogeneity and survival of the fittest cells. The strong plasticity of cancer cells leads to the rewiring of signaling pathways and metabolic networks, all in response to changes of their micro-environmental conditions. For all these reasons, mathematical modeling of cancer evolution should include several biological scales: molecular, cellular and tissular. In this project we represent cancer cells as distributions over spatial and multiple metabolic dimensions and study their evolution using high dimensional partial differential equations models.

Machine learning and image analysis background

Various AI – including machine and deep learning – methods will be used to quantify the tumor spatial heterogeneity at different scales (distribution of blood vessels, invasion fronts, cell clusters and distribution of metabolic markers) from H&E-stained histological sections and from mass cytometry tissue imaging datasets. Ultimately, these methods will feed the mathematical model with information (initial data and parameters) needed for predicting tumor evolution upon targeted therapy. Focused tasks, involving our biomedical partners, will be dedicated to the explicability of the innovative AI approaches, allowing us to develop a creative framework.

Specific MALMO internship projects

1) Mechanistic models of plasticity in melanoma submitted to targeted treatment

The most frequent mutations in melanoma affect the BRAF oncogene, a protein kinase of the MAPK signaling pathway. Therapies targeting both BRAF and MEK are effective for only 50% of patients and, almost systematically, generate drug resistance. Genetic and non-genetic mechanisms associated with the strong heterogeneity and plasticity of melanoma cells have been suggested to favor drug resistance but are still poorly understood. In this internship we will use changes of the transcriptional program of tumors (bulk and single cell RNA-seq data) submitted to treatment to build mechanistic models of tumor heterogeneity and plasticity. Our goal is to characterize pathways that contribute to resistance to BRAF and MEK inhibitors of melanoma and build mechanistic models that are consistent with the available data. In order to obtain these models we will use hierarchical modeling and learning: i) models of different types (from qualitative networks to quantitative differential equations models) will be placed in an hierarchy and connected one to another by mapping of components and parameters ii) in the machine learning process, simpler models will constrain more complex models. The intern will extract the data from public databases, will analyze the changes of transcription programs induced by treatment using data dimensionality reduction methods, and will build the hierarchical models needed for the project. The mechanistic simplified network will be integrated in a mesoscale PDE model whose predictions will be compared to our previous phenomenological model [1].

Prerequisites: Basic network and graph modeling, good knowledge of ordinary differential equations and partial differential equations modeling, basic knowledge of machine learning, excellent coding skills in Python and/or Matlab. Basic knowledge of biochemistry, signaling and metabolic pathways would be appreciated. 

This internship may continue with a PhD thesis on mathematical modeling of spatial biology of cancer.


  1. A Hodgkinson, D Trucu, M Lacroix, L Le Cam, O Radulescu. Computational Model of Heterogeneity in Melanoma : Designing Therapies and Predicting Outcomes. Frontiers in Oncology, 12 (2022).
  2. G. Anandalingam and T.L.Friesz. Hierarchical optimization: An introduction. Annals of Operations Research 34, 1-11 (1992).

To apply: send CV + motivation letter + transcripts to

2) Simulating high dimensional models of tumor metabolic plasticity by probabilistic methods and deep learning

Tumor cell dynamics is described by high dimensional partial differential equations. Indeed, tumor cells are characterized by their position in the physical space, but also by many other dimensions characterizing the functioning of the intracellular signaling and metabolic pathways. This raises important challenges concerning the computational cost, the precision and stability of traditional numerical simulation schemes based on finite differences or elements. In this internship we will investigate two alternative approaches. The first approach is probabilistic and based on Monte Carlo simulations. By the Feynman-Kac formula, solutions of high dimensional reaction-diffusion PDEs can be related to solutions of high dimensional stochastic differential equations. The advantage of the probabilistic approach over finite differences or elements is that the convergence rate does not depend on the dimension of the problem and suffers less from the curse of dimensionality. The second approach exploits the universal approximation properties of artificial neural networks (ANN). ANNs have the capacity to overcome the curse of dimensionality: they can approximate solutions of PDEs arbitrarily well with parametric complexity growing polynomially in the PDE dimension. The methods will be implemented using Python, Julia or Matlab.

Prerequisites: Good theoretical knowledge of partial differential equations and stochastic differential equations, familiarity with  IA (deep learning), excellent coding skills in Python and/or Julia/Matlab.

This internship may continue with a PhD thesis on high dimensional PDEs or on modeling the spatial biology of cancer.


  1. A Hodgkinson, D Trucu, M Lacroix, L Le Cam, O Radulescu. Computational Model of Heterogeneity
    in Melanoma : Designing Therapies and Predicting Outcomes. Frontiers in
    Oncology, 12 (2022).
  2. E Weinan, J Han, A Jentzen. Algorithms for solving high dimensional PDEs: from nonlinear Monte Carlo to machine learning. Nonlinearity 35, 278 (2022).
  3. M. Hutzenthaler, A. Jentzen, T.Kruse, TA Nguyen. A proof that rectified deep neural networks overcome the curse of dimensionality in the numerical equation of semilinear heat equations. SN
    Partial Differential Equations and Applications, 1, 10 (2020), arXiv:13333v2.

To apply: send CV + motivation letter + transcripts to


Medical software is an essential component in clinical tool kits. They are designed to expedite the time in which clinicians can run their diagnostic and prognostic processes in a more repeatable and reproducible manner (i.e., minimize the occurrence of epistemic uncertainty). However, not all software available is completely automated; many medical applications are still reliant on semi-automated processes. This is particularly evident in digital pathology, in which a pathologist spends considerable time manually evaluating Whole Slide Images (WSIs) – digitized forms of microscopic slides. In the MALMO project, we are completely automating the diagnostic and prognostic process so that pathologists can extract relevant clinical information with the click of a button. The software in its entirety is developed using the programming language Python, and provides clinicians with both 2D and 3D analyses. However, there are some limitations with utilizing Python for 3D rendering. One of these limitations include the speed of interaction with a Python-rendered 3D model (i.e., there can be a time delay when interacting with 3D models). Furthermore, a clinician should be able to easily interact with a 3D model, including zooming in and out of regions of interest that may have clinical significance, and this is a feature not readily achievable in Python. However, Python does not need to be used in isolation; it can be coupled with a 3D computer graphics software to achieve the ultimate 3D model. In this internship, you will assist in combining our current Python pipeline with the open-source software Blender to develop more interactive and intuitive 3D models. The Blender-based 3D rendering will need to be connected to the Python-based GUI. This coupling will ensure that end-users have a simple and easy-to-use GUI interface that will require very little training for the pathologist to use. But the simple interface will have the benefit of being connected with the powerful Blender software to produce high quality and completely animated 3D visualization. The speed and efficiency of the 3D rendering is essential, as the pathologists will need to be able to use this 3D model in a real-time clinical setting. An option of cloud deployment of this module will be studied, towards the end of the internship.



Traditionally, software has been made available to clinicians by allowing them to download the package and install it onto a local machine. However, previously, software were more static and updates may have been more infrequent. Today, several deep learning (Python, PyTorch) modules are used for digital pathology that will need to be made readily available to clinicians. Furthermore, these deep learning modules will continually evolve as new data becomes available. Dynamic AI-based software is not conducive to the traditional format of downloading and installing software. Furthermore, the traditional format also required the availability of technical support (e.g., IT), which is not a resource currently available in the MALMO project. To make our software readily available with minimal fuss for the clinician, and allow the software to automatically update without little to no intervention required from the clinician or even from a technical support, an alternative solution would be to make the software readily available via a Cloud service, such as AWS, GCP, OVH cloud. This will ensure pathologists can easily access the software and require very little technical support (no installation is needed since the software is working within a browser). For this internship, you will be responsible for deploying and ensuring the complete functionality of our software on a cloud-based server. You will identify the most appropriate Cloud service for the project, develop a pipeline for the Python- and Blender-based softwares deployment on the Cloud, identify the needs and requirements of maintenance on the Cloud, and identify ways to protect the software on the Cloud (e.g., using differential privacy).



LPHI1, Monpellier, is an innovative and high standard laboratory for basic research in biology. It hosts the Computational Systems Biology Team (CSBT), one of few of this kind in France, developing projects at the interface between Biology, Physics and Mathematics. Team leader of CSBT, Ovidiu Radulescu is with University of Montpellier. Established in 1289, the University of Montpellier (UM) is the 6th largest university in France, with about 50,000 students including 7000 foreign students. One of the most innovative higher education institutions in the world, UM ranks very high in many international rankings : first in the world in the 2018 Shanghai ranking for Ecology, first most innovative French university in 2018 Reuter’s ranking, 5th in France in 2018 Leiden’s ranking for the quality of its scientific publications, 3rd French university in the 2019 “University Impact ranking” of Times Higher Education. These increasingly outstanding results reflect the dynamism triggered by the Montpellier University of Excellence I-SITE project since the prestigious certification was obtained in March 2017. Montpellier is a vibrant and sunny Southern France city. It benefits of the Mediterranean coast and proximity of the Cevennes mountain range, has a beautiful old city centre and great infrastructures.

The “Institut de Recherche en Cancérologie de Montpellier” (IRCM), is embedded in Montpellier Cancer Center. This research institute has raised its research to the highest international level in the field of fundamental and applied oncology. Research is carried out in close collaboration with the clinical departments of the Centre de Lutte Contre le Cancer de Montpellier (ICM: Institut du Cancer de Montpellier), and industrial partners. Jointly operated by Inserm, ICM and the University of Montpellier, the IRCM now brings together more than 200 people, researchers, clinicians, technicians and students, organized into 17 research teams supported by efficient core facilities and support services, including an innovative mass cytometry imaging platform that will be central to our project. In an extremely competitive and rapidly evolving field of research, the objectives of IRCM is to accelerate innovation and transfer new discoveries to the clinic.

Paris Brain Institute2 (ICM – Institut du Cerveau) – CNRS, Inserm, Sorbonne, AP-HP, INRIA team « Aramis ». Located within the Pitié-Salpêtrière hospital, Paris, ICM is an international research center whose innovative concept and structure make it unique. The best scientists from all backgrounds and countries come together at the Institute to perform leading-edge research in this area. Daniel Racoceanu is Principal Investigator @ ICM and Professor at Sorbonne University, a multidisciplinary, research-intensive, world-class university. Located in the heart of Paris, with a regional presence, this university is committed to the success of its students and to meeting the scientific challenges of the 21st century. Thanks to its 55,300 students, 6,400 academic researchers and partner researchers, and 3,600 administrative and technical staff who make it a daily reality, Sorbonne University promotes diversity, creativity, innovation and openness to the world.

2 Paris Brain Institute (ICM) :