Memorial session in honor of Ezio Venturino
Stability, oscillations, carrying capacities, infections and pattern formation in population-dynamical systems
Horst Malchow
The lecture is dedicated to the memory and appreciation of Ezio Venturino. After a few words about our friendship and collaboration, brief examples of the results of our joint work will be presented. These come from the model dynamics of predator-prey systems with infection or age classes of prey.
Dispersal in discrete-time population models
Daniel Franco
Many species live in fragmented habitats, which makes it important to understand how movement between regions affects them. One key question is how dispersal influences the overall population size. In this talk, we will share recent findings showing how the total population can change as the dispersal rate increases, depending on the local dynamics, the metapopulation topology or fluctuations in the environment.
Eco-epidemiological models with trait- and density-mediated effects of parasitized predators
Frank M. Hilker
We consider the case of an infectious disease circulating in a predator population. The infection may exert density-mediated effects on the predators (e.g., increasing mortality or decreasing fecundity) or trait-mediated effects (e.g., changing behavior, physiology, or morphology). Little is known about trait-mediated effects in eco-epidemiological systems, both empirically and theoretically. Here, we assume that infection alters the attack rate and the conversion efficiency of predators. We find that trait-mediated effects are always negative for predators and can be positive or negative for prey. Furthermore, infection-induced increases in the attack rate can enable the survival of predators which would go extinct in the absence of trait-mediated effects. Mathematically, the latter occurs via transcritical bifurcations or saddle-node bifurcations (for frequency- and density-dependent disease transmission, respectively). The results underline the importance of trait-mediated effects and provide fundamental insights from generic eco-epidemiological models. This is joint work with the late Ezio Venturino.
The role of manipulative trophically transmitted parasites in the stability of a predator–prey community
Jean-Christophe Poggiale
In this work, we are interested in the effect of a trophically transmitted parasite on the structure and dynamics of a resident predator–prey community. The parasite, apart from increasing the mortality rates of its hosts, can also change their physiology or their behaviour, which is known as trait-mediated indirect interaction. We assume that parasite transmission, which entails rapid physiological or behavioural change, is a fast process with respect to the community dynamics,
including prey and predator growths and predation. This existence of different time scales allows us to provide analytical results to understand some conditions under which the parasite change the dynamics of the predator–prey community. Thus, we are able to find conditions under which indirect trait-mediated interactions induced by the parasite lead to a coexistence between predators and prey that would not occur in its absence. We also provide conditions, evolutionary deleterious, that ensure the extinction of a predator–prey community that would be viable without parasite intervention. Finally, we show situations in which the action of the parasite destabilizes the predator–prey system without eliminating it, producing oscillations, the mechanism of which is analysed.
Ecology, Evolution and Population Dynamics
What process(es) explain(s) genetics diversity in the wild? Applying mathematical models to study biological questions
Guillaume Achaz
Neutral theory (i.e. Wright-Fisher like models) assumes that in a population of size N, diversity results from an equilibrium between new mutations arising at rate μ and genetic drift that purge them at rate 1/N, predicting an equilibrium value proportional to Nμ. The difference between this expectation and the much lower observed molecular diversity is known as the Lewontin’s paradox of variation. Here, I will discuss how different mathematical evolutionary models were tailored to address this paradox of variation and more specifically why a model of genetic draft, a regime of evolution where recurrent sparse selective sweeps entirely drive the diversity of complete genomes is an excellent candidate to solve the paradox. I shall then display few open problems regarding the derivation and the characterization of these models.
Impact of complex spatial population structures on biological evolution
Anne-Florence Bitbol
Stochastic models are very useful to describe biological evolution, in particular because fluctuations due to small numbers of individuals cannot be ignored. Indeed, a mutation usually appears in a single individual within a population. Microbial populations often have complex spatial structures, with homogeneous competition holding only at a local scale. Population structure can strongly impact evolution, in particular by affecting the probability that mutations take over. I will first present a general model for describing structured populations on graphs. By tuning migration asymmetry in the rare migration regime, the star graph transitions from amplifying to suppressing natural selection. However, suppression of selection is pervasive in the regime of frequent migrations. Then, I will discuss how these results are impacted by environment heterogeneity across demes. Finally, I will show that spatial structure can foster the evolution of drug resistance.
Trait evolution in a Moran Model: Impact of environment and population structure
Hélène Leman
In this talk, I will introduce a population model based on the Moran model, which includes individual competition, mutation, environmental changes, and spatial structure. We will explore various parameter scalings, first examining a scenario with rare mutations and then a scenario with mutations having minimal effects. In each case, we will consider different rates of environmental changes. Then, for a constant environment, we will investigate the influence of the population's spatial structure. Specifically, if the population is organized into demes, we will analyze the dynamics within a deme when the number of demes is infinite. These different limits yield various forms of canonical equations that we can compare. This is a joint work with A. Lambert, H. Morlon, J. Tchouanti and T. Vo.
Loose linkage in the ancestral recombination graph
Frederic Alberti
Understanding the interplay between recombination and resampling is a significant challenge in mathematical population genetics. Asymptotic results about the distribution of samples when recombination is strong compared to resampling are often based on the approximate solution of certain recursions, which is technically challenging. This work generalises an elegant probabilistic argument, based on the coupling of ancestral processes, thus far only available in the case of two sites, to the multilocus setting. In doing so, we are able to obtain additional conceptual insight into, and slightly generalise, a classical result on closed-form asymptotic sampling distributions by A. Bhaskar and Y. Song.
Contributed talks for the open session
Hematopoiesis as a continuum: from stochastic compartmental model to hydrodynamic limit
Fernández Baranda Ana
A general PDE for travelling waves and adaptive dynamics
Ged François
How Cross-Feeding Shapes Stability and Persistence in Microbial Communities
Faul Louis
ToMATo clustering algorithm for neuronal spike sorting
Martineau Louise
AI for Health Science
AI-Driven Healthcare Management: From Emergency Resource Optimization to Crisis Detection
Stephan Robert-Nicoud, Léonard Truscello
This presentation will highlight the use of artificial intelligence for healthcare management. A first study examined the optimization of emergency medical resources, particularly ambulance allocation and relocation. In recent decades, these services have faced increasing workload due to population aging and the pandemic, leading to longer response times and strained capacities. Reinforcement learning methods were developed but showed that external factors such as traffic or weather had little influence. Instead, time-of-day effects dominated, which motivated the design of stochastic models based on inhomogeneous Poisson processes, offering effective and transparent decision support. Building on these results, new perspectives are considered in the context of exceptional health crises such as pandemics. The emergence of such crises is often first visible in press and social media streams, but these, along with medical datasets, present major challenges due to their noisy, imbalanced, or imperfectly labeled nature. Current exploratory work investigates the potential of stochastic label perturbation (DisturbLabel), an underexplored regularization method, as a way to improve robustness in medical and textual datasets.
Health Data and AI: Application to HIV
Jennifer Lagoutte-Renosi, Marie-Blanche Valnet Rabier, Ornella Jutcha Nganwa
The therapeutic management of patients living with HIV is guided by national and European recommendations in order to limit transmission of the virus. However, antiretroviral treatments must be taken continuously in order to be effective. Patient compliance is therefore crucial in limiting viral resistance and transmission. Compliance depends on the acceptability of the dosage forms and the occurrence of adverse effects. Currently, there is no algorithm that can predict the risk of adverse effects from antiretroviral drugs. The OCTAVIA project aims to develop a diagnostic tool for clinicians to assess the risk of adverse effects in order to improve patient compliance. The project is based on real-life data on adverse effects from the global pharmacovigilance database, Vigibase, and the French clinical database of patients living with HIV, NADIS. The use of these two databases should make it possible to identify profiles of patients at risk of adverse effects. The OCTAVIA tool will ultimately enable patients to be classified according to their risk of adverse effects, taking into account co-treatments, co-morbidities, aging, and other biological characteristics.
Graphing the spread: Using graph neural networks to predict multi-drug resistant infection risks in hospitals
Douglas Teodoro
Healthcare-associated infections (HAIs), particularly those caused by multidrug-resistant (MDR) organisms, present a significant burden on global healthcare systems, leading to increased patient mortality, longer hospital stays, and substantial financial costs. While Infection Prevention and Control (IPC) programs are critical in reducing HAIs, identifying patients at high risk remains a challenge in surveillance programs. This presentation outlines a novel approach that treats infection spread as a network problem, using Graph Neural Networks (GNNs) to predict MDR infection risks by modeling the complex interactions within a hospital setting.
Two distinct GNN-based models are presented. The first model uses a static graph to represent interactions between patients in an intensive care unit, with connections (edges) defined by shared wards or healthcare workers. Trained on the publicly available MIMIC-III dataset to predict MDR Enterobacteriaceae colonization, this model achieved a high predictive accuracy, with an Area Under the Receiver Operating Characteristic Curve (AUROC) of up to 96%. Recognizing that hospital environments are dynamic, a more advanced model, the Space-Time-and-Memory GNN (STM-GNN), was developed to capture temporal changes in patient and environmental networks. This model incorporates spatial, temporal, and memory components to learn from evolving contact patterns. Evaluated on a unique dataset containing both clinical and environmental samples from a long-term care facility, the STM-GNN outperformed traditional machine learning and other GNN baselines, achieving an AUROC of 0.84 in a setting with sparse data.
Together, these results demonstrate that GNNs are a powerful tool for proactive, data-driven infection prevention. By learning from the inherent network structure of patient contacts, this approach can enhance IPC surveillance and support the early identification of patients at risk of colonization by MDR pathogens.
AI-Powered Analysis of Cardiovascular Signals for the Prevention and Management of Cardiovascular Diseases
Mohamed Sraitih
The development of automatic systems for cardiovascular disease detection must meet several essential criteria to ensure clinical applicability and effectiveness. These include high reliability, low computational complexity, and robust decision-making performance—attributes that remain critical in real-world healthcare environments. This study presents a recent deep learning methodologies employed for denoising electrocardiogram (ECG) signals and automating cardiovascular disease detection.
Multi-objective approaches to the home health care problem
Jérémy Decerle
The home health care problem involves the complex routing and scheduling of caregivers who provide essential services to patients at home. To address the growing demand and operational challenges, we study a multi-objective version of this problem that integrates skills, time window, and synchronization constraints while balancing several conflicting goals: minimizing total working times, maximizing service quality, and ensuring workload fairness among caregivers. In this presentation, we introduce two complementary approaches to solve this problem. First, a hybrid algorithm that combines memetic and ant-colony optimization principles. Second, a purely multi-objective memetic algorithm designed to explicitly explore trade-offs among efficiency, fairness, and service quality. Computational experiments on benchmark instances from the literature show both methods outperform several existing metaheuristics and a commercial solver. Particular attention will also be brought to the trade-offs between objectives highlighting implications for realistic, applicable home health care planning.
