HAL Evry
Not a member yet
19237 research outputs found
Sort by
Self-supervised Representation Learning on Gene Expression Data
No full textInternational audienceMotivation Predicting phenotypes from gene expression data is a crucial task in biomedical research, enabling insights into disease mechanisms, drug responses, and personalized medicine. Traditional machine learning and deep learning rely on supervised learning, which requires large quantities of labeled data that are costly and time-consuming to obtain in the case of gene expression data. Self-supervised learning has recently emerged as a promising approach to overcome these limitations by extracting information directly from the structure of unlabeled data. Results In this study, we investigate the application of state-of-the-art self-supervised learning methods to bulk gene expression data for phenotype prediction. We selected three self-supervised methods, based on different approaches, to assess their ability to exploit the inherent structure of the data and to generate qualitative representations which can be used for downstream predictive tasks. By using several publicly available gene expression datasets, we demonstrate how the selected methods can effectively capture complex information and improve phenotype prediction accuracy. The results obtained show that self-supervised learning methods can outperform traditional supervised models besides offering significant advantage by reducing the dependency on annotated data. We provide a comprehensive analysis of the performance of each method by highlighting their strengths and limitations. We also provide recommendations for using these methods depending on the case under study. Finally, we outline future research directions to enhance the application of self-supervised learning in the field of gene expression data analysis. This study is the first work that deals with bulk RNA-Seq data and self-supervised learning. Availability The code and results are available at https://github.com/kdradjat/ssrl-rnaseq. Supplementary information Supplementary data are available at Bioinformatics online
Variable Threshold-Oriented Event-Triggered Cluster Consensus for Groups of Multiagent Systems
No full textInternational audienceThis article investigates the leader–following cluster consensus for generic linear heterogeneous multiagent systems (MASs). Unlike the existing research, a novel event-triggered (ET) control mechanism is designed and developed on the transmission side of the agents, over directed communication topologies, to reduce communication load. For this purpose, a variable threshold function as the fully distributed ET condition (ETC) is suggested, which provides a smooth transition and considers both maximum and minimum threshold levels for triggering. A relative-state feedback-based cluster consensus control protocol is designed by considering the cooperative and competitive interaction behavior of agents. Then, the convergence analysis is performed by utilizing the Lyapunov method. This work is then further extended for the ET observer-based output feedback cluster consensus problem. The proposed ETC naturally eliminates the Zeno behavior for each agent. In contrast to existing methods, a variable threshold-based ET scheme, a cooperation-competition network, and an elimination of Zeno behavior for both state-based and output-based methods have been considered for the leader–following cluster consensus. Finally, illustrative examples are used to validate the theoretical results
An orthogonal data-driven Neural Network observer for simultaneous state and unknown input estimation for Linear systems
No full textInternational audienceThis paper presents a novel approach for state and unknown input (UI) observer in single-input single-output (SISO) linear time-invariant (LTI) systems. An Orthogonal Neural Network (ONN), which does not require traditional training or offline learning, is incorporated to approximate the unknown input. The proposed method extends the system state to include the ONN weights, creating a new LTI system. Leveraging the structure of orthogonal activation functions, a suitable algorithm for the extended state dynamics is derived. The observer gain is computed using Linear Matrix Inequalities (LMI), ensuring stability and attenuating the approximation error. Unlike existing methods, this approach requires no assumptions about the unknown input, such as matching conditions or boundedness. A numerical example demonstrates the effectiveness of the proposed scheme
Nanopores solides pour améliorer la détection des biomolécules : contrôle des forces motrices et de la dynamique
No full textSolid-state nanopores have shown great potential in biomolecular analysis, medical diagnostics, and environmental monitoring in recent years. Up to now these artificial materials are under used for biological applications. I investigated the dynamics of biomolecules in crowded and confined environments. I first introduce the physical background of the dynamics of polymer in dilute and semi-dilute solution in confined medium. These physical frameworks provide the basis for understanding how crowding conditions influence molecular diffusion, translocation rates, and structural behavior. In the second part, I describe the experimental methods used during my work. In the third part, I demonstrate a universal behavior governing the entry of large polymer chains into nanopores by electrical detection at the single-molecule scale. Specifically, when the mesh size of the semi-dilute polymer solution is smaller than the nanopore diameter, the osmotic pressure generated by the polymer chains can overcome the entropic cost of confinement, allowing polymer chains to enter the nanopore. We determine the threshold concentration required for polymer entry and find that it exhibits a power-law dependence on the ratio between nanopore size and monomer size, which agrees with the theoretical prediction by Pierre-Gilles de Gennes. Furthermore, I show that the depletion layer thickness at the nanopore entrance scales with the screening length (mesh size) of the solution, and I identify three typical translocation regimes depending on the ratio between nanopore diameter and screening length. In the fourth part, I investigate DNA stability in the presence of polyamines. DNA compaction and decompaction are involved in gene expression regulation. Polyamines are involved in physiological functions of the cell but also for several human diseases. We show in the presence of L-DOPA on the solid-state nanopore surface that we screen non-specific interaction between the DNA and the nanopore. Then we decrease the energy barrier for DNA entry into the nanopore. In these conditions we control the dynamics. We show the DNA aggregation and resolubilization using solid-state nanopore as a function of spermidine (positive charge 3+) and spermine (4+) concentration. We obtain the same threshold for the DNA precipitation and resolubilization in bulk measurements and from single-nanopore recordings. At high salt concentration we prevent any aggregation of DNA with polyamines. We discussed these results according polyelectrolyte behavior. In summary, my work demonstrates that combining controllable crowding environments with solid-state nanopore technology provides a powerful strategy for probing biomolecular dynamics at the single-molecule level. The findings offer broad applications in biosensing, polymer physics, and the study of phase behavior in confined systems.Les nanopores solides ont montré un grand potentiel dans l'analyse des biomolécules, le diagnostic médical et la surveillance environnementale au cours des dernières années. Jusqu'à présent, ces matériaux artificiels sont sous-utilisés pour des applications biologiques. J'ai étudié la dynamique des biomolécules dans des environnements encombrés et confinés. Je présente d'abord le contexte physique de la dynamique des polymères en solution diluée et semi-diluée dans un milieu confiné. Ces cadres physiques fournissent la base pour comprendre comment les conditions d'encombrement influencent la diffusion moléculaire, les vitesses de translocation et le comportement structurel. Dans la deuxième partie, je décris les méthodes expérimentales utilisées au cours de mon travail. Dans la troisième partie, je démontre un comportement universel régissant l'entrée de longues chaînes de polymères dans les nanopores, par détection électrique à l'échelle de la molécule unique. Plus précisément, lorsque la taille des mailles de la solution polymère semi-diluée est inférieure au diamètre du nanopore, la pression osmotique générée par les chaînes de polymères peut surmonter le coût entropique du confinement, permettant aux chaînes de polymères d'entrer dans le nanopore. Nous déterminons la concentration seuil nécessaire à l'entrée des polymères et constatons qu'elle présente une dépendance en loi de puissance par rapport au ratio entre la taille du nanopore et la taille du monomère, ce qui est en accord avec la prédiction théorique de Pierre-Gilles de Gennes. En outre, je montre que l'épaisseur de la couche de déplétion à l'entrée du nanopore est proportionnelle à la longueur d'écrantage (taille des mailles) de la solution, et j'identifie trois régimes de translocation en fonction du rapport entre le diamètre du nanopore et la longueur d'écrantage. Dans la quatrième partie, j'étudie la stabilité de l'ADN en présence de polyamines. La compaction et la décompaction de l'ADN sont impliquées dans la régulation de l'expression des gènes. Les polyamines sont impliquées dans les fonctions physiologiques de la cellule, mais aussi dans plusieurs maladies humaines. Nous montrons qu'en présence de L-DOPA à la surface du nanopore solide, nous réduisons les interactions non spécifiques entre l'ADN et le nanopore. Ensuite, nous diminuons la barrière énergétique pour l'entrée de l'ADN dans le nanopore. Dans ces conditions, nous contrôlons la dynamique. Nous montrons l'agrégation et la redissolution de l'ADN à l'aide de nanopores solides en fonction de la concentration en spermidine (charge positive 3+) et en spermine (4+). Nous obtenons le même seuil pour la précipitation et la redissolution de l'ADN dans les mesures en volume et dans les enregistrements à nanopore unique. À forte concentration en sel, nous empêchons toute agrégation de l'ADN avec les polyamines. Nous avons discuté ces résultats selon le comportement des polyélectrolytes. En résumé, mon travail démontre que la combinaison d'environnements d'encombrement contrôlables avec la technologie des nanopores solides constitue une stratégie puissante pour sonder la dynamique des biomolécules à l'échelle de la molécule unique. Les résultats offrent de larges applications dans la biosélection, la physique des polymères et l'étude du comportement de phase en milieu confiné
Beyond overweight, visceral adiposity is associated with estimation of cardiovascular risk in patients living with type 1 diabetes: findings from the SFDT1 cohort
No full textConflicts of interests: EBP reports receiving lecture honorariums from Astra Zeneca and Sanofi, and has been an employee of Boehringer Ingelheim since February 2024International audienceIntroduction & Objectives As in the general population, people living with type 1 diabetes (PWT1D) are faced with overweight and obesity, which contribute to cardiovascular (CV) risk. However, the role of visceral adiposity, due to its adverse metabolic profile, should also be addressed in PWT1D. We aimed to assess the 10-year CV risk of PWT1D according to body mass index (BMI) and waist-to-height ratio (WHtR), a parameter for estimating visceral adiposity. MethodsIn this cross-sectional study, PWT1D in primary CV prevention from the SFDT1 cohort were categorized by BMI status, either normal (18.5-24.9 kg/m 2 ) or overweight/obesity (≥ 25 kg/m 2 ), and by WHtR according to the validated threshold of 0.5. The 10-year CV risk was estimated using the Steno Type 1 Risk Engine and classified into three categories: low (< 10%), intermediate (10-20%) and high (> 20%). The distribution of CV risk was assessed using density plots. In multivariable analysis, the association between BMI, WHtR, and high estimated 10-year CV risk was studied using spline regression models with sex stratification. Thresholds were determined by the Receiver Operating Characteristic (ROC) curve. ResultsThe study included 1,482 patients; 49.9% had a normal BMI, and 50.1% a BMI ≥ 25 kg/m 2 . The proportion of patients with high CV risk was higher in PWT1D with overweight/obesity (12% vs. 7%) and in those with WHtR ≥ 0.5 (13% vs. 4%). BMI was significantly associated with high CV risk in men (p = 0.001) but a non-significant trend was found in women (p = 0.053). WHtR was significantly associated with high CV risk in both men (p < 0.001) and women (p = 0.046). The BMI threshold associated with high CV risk was 24.9 kg/m 2 for men, and the WHtR threshold was 0.5 for both men and women. ConclusionIn PWT1D in condition of primary CV prevention, visceral adiposity, assessed by WHtR, is a more robust marker of estimated 10-year CV risk than overweight/obesity status in both men and women.</div
Antisense-mediated regulation of nitric oxide biosynthesis
No full textInternational audienceNitric oxide (NO) is a gasotransmitter-diffusible signaling molecule that is critical across organisms. In plants, NO plays an important function in growth and development, senescence, and the response to abiotic and biotic stress (Domingos et al., 2015). Our understanding of the mechanism for NO biosynthesis in plants is limited and centered around the nitrate reductase pathway. Two known NR-encoding genes in Arabidopsis thaliana, NIA1/2, are active in leaves and meristematic tissue (Olas and Wahl, 2019), independently regulated in a tissue-dependent manner (Vidal et al., 2015), and required for NO synthesis (Chen et al., 2016). Non-coding RNA species, i.e., antisense, long intergenic, and sequences processed into small microRNAs/small interfering RNAs (siRNAs) can be potent regulators of gene expression in plants (Traubenik et al., 2024). Previous work discovered two non-coding species originating from the NIA1/2 loci including a 22-nt siRNA (Wu et al., 2020) as well as an antisense RNA (asRNA) from the 3′ UTR (Chekanova et al., 2007). The small siRNA derived from the NIA1/2 loci was found to restrain NIA1/2 translation, ultimately inhibiting plant growth and enhancing the stress response (Wu et al., 2020). How these long non-coding species specifically, as NIA1/2, regulate their respective protein-coding sense strands was not understood. In mammalian studies, a similar type of induced NO synthase (iNOS) protein has been shown to be regulated by its asRNA transcribed from the 3′ UTR. It was determined that a iNOS mRNA–asRNA–protein complex is formed with embryonic lethal abnormal visual-like (HuR) and AU-binding factor 1/heterogeneous nuclear ribonucleoprotein L, inhibiting degradation and ultimately stabilizing iNOS mRNA (Galbán et al., 2008; Matsui et al., 2008). Recently, Li et al. (2025) propose a new mechanism by which NO biosynthesis is regulated via an antisense–mRNA–protein complex, itself also regulated via mRNA modification, and a subsequent secondary structure to control stomatal movement (Figure 1)
6-DOF Pose Estimation For Event Cameras Using A Transformer-Based Approach
No full textInternational audienceEvent cameras are novel sensors that provide significant advantages over traditional cameras, such as low latency,high dynamic range, and reduced motion blur. These properties make them particularly well-suited for 6-DOF poseestimation tasks in challenging environments. In this paper, we present a novel transformer-based approach for 6-DOF pose estimation using event camera data. Our method combines a pretrained ResNet50 backbone for featureextraction with a custom transformer encoder to model the spatial and temporal dependencies inherent in eventdata. We demonstrate the effectiveness of our approach on a dataset of real-world event camera images, where weachieve significant improvements in pose estimation accuracy compared to state-of-the-art methods. Additionally,our method exhibits robustness to varying lighting conditions, motion blur, and sensor noise, highlighting itspotential for deployment in a wide range of applications, such as robotics, autonomous vehicles, and augmentedreality. Our experimental results showcase the promising capabilities of transformer-based models in leveragingthe unique properties of event cameras for accurate and efficient 6-DOF pose estimation
Simulation Immersive pour la Formation des forces de l'ordre : Enrichir la Gestion de Conflits Interpersonnels par l'Intégration du Toucher en Réalité Virtuelle
No full textInternational audienceIntegrating tactile feedback into virtual reality simulators offers new opportunities for professional training, especially in fields involving interpersonal conflict management. This article presents an immersive approach that combines simulated social interactions and haptic sensations, applied to a realistic intervention scenario.L’intégration du retour haptique dans les simulateurs de réalité virtuelle ouvre de nouvelles perspectives pour la formation professionnelle, notamment dans les métiers impliquant la gestion de conflits interpersonnels. Cet article présente une approche immersive qui combine interactions sociales simulées et sensations haptiques, appliquée à un scénario réaliste d’intervention
Comparison of Machine Learning and the ELK Stack for Automated Cyber Attack Detection: A Comparative Approach
No full textInternational audienceThe increasing complexity of cyberattacks has heightened the demand for effective detection mechanisms beyond traditional SIEM systems, which still depend on manual analysis. This study evaluates common classification algorithms—Decision Tree (DT), Random Forest (RF), Support Vector Machine (SVM), and Naive Bayes (NB)—and addresses a key limitation in existing datasets: the lack of detail on attack techniques. To overcome this, we simulated nine attack techniques and applied both manual and automated detection approaches using the ELK Stack and Artificial Intelligence (AI) models. While the manual method enhanced system visualization, it yielded accuracy below 39%. In contrast, automated classification achieved near-perfect results: DT and NB reached 100% accuracy, RF and SVM 99% in multi-class tasks; for binary classification, DT, RF, and NB reached 100%, with SVM at 99.98%. NB proved the most efficient in balancing accuracy, scalability, and resource use