Mines Paris, PSL University

Structural geological modeling is aimed at finding a representation of geological units. This is a complex ill-posed problem, and the data may be sparse and of varying quality, leading to multiple geological models consistent with them. Despite continuous advances for decades in geological modeling, recent studies still show some unresolved industrial challenges. Consequently, this work aims to improve geological modeling by proposing a novel approach with a focus on improving uncertainty management. A stochastic approach is developed based on deep generative methods, namely generative adversarial networks (GANs). Thanks to a synthetic dataset, a GAN is trained to generate two-dimensional unconditional geomodels that are plausible. Subsequently, a Bayesian inversion is performed with a Metropolis-adjusted Langevin algorithm (MALA) to produce geomodels consistent with field data. The proposed approach is validated on different conditioning data. For each case, the approach is able to successfully produce a variety of geomodels that can be linked to different geological settings. The uncertainties on geological units are measured by Shannon entropy on the generated models.

Contemporary formal pragmatics has uncovered a rich inferential typology in spoken languages, one that includes at-issue contents, presuppositions, implicatures, homogeneity inferences, supplements, and expressives. The division of informational contents among this typology is sometimes taken to be specified in the lexicon, but gestural research has argued against this view: with one possible exception (expressives), participants productively divide the content of iconic representations among the slots of the inferential typology, and this can be shown with pro-speech (= word-replacing) gestures, and with novel visual animations. Despite important recent developments, sign language semantics has not systematically investigated this inferential typology. Based on published and new data from ASL, we do so from a dual perspective: we exhibit the characteristic behavior of different inferential types in lexical signs, but also (when applicable) in iconically modulated constructions, notably classifier predicates. These have a lexically specified form, but an entirely free position and movement in signing space, which are interpreted iconically and give rise to truth conditions that couldn’t be stored lexically. Classifier predicates thus make it possible to replicate the productivity argument from pro-speech gestures and visual animations, but with greater ease and precision, because unlike pro-speech elements, they are a common and fully integrated part of sign language. They also address an objection to findings coming from pro-speech gestures and visual animations, namely that these are in essence codes for normal expressions of spoken language (words); this objection has no plausibility for ASL iconic constructions, as these are normal expressions of sign language. Besides highlighting the importance of sign language for formal pragmatics, our study makes a broader point: any analysis of sign language must provide an explicit treatment of its iconic component and of its interaction with the inferential typology.

To address the urgent biodiversity crisis, it is crucial to understand the nature of plant assemblages. The distribution of plant species is not only shaped by their broad environmental requirements, but also by micro-environmental conditions, dispersal limitations, and direct and indirect species interactions. While predicting species composition and habitat identity is essential for conservation and restoration purposes, it thus remains challenging. In this study, we propose a novel approach inspired by advances in large language models to learn the "syntax" of abundance-ordered plant species sequences in communities. Our method, which captures latent associations between species across diverse ecosystems, can be fine-tuned for diverse tasks. In particular, we show that our methodology is able to outperform other approaches to (i) predict species that might occur in an assemblage given the other listed species, despite being originally missing in the species list (+16.53% compared to co-occurrence matrices and +6.56% compared to neural networks) and (ii) classify habitat types from species assemblages (+5.54% compared to expert systems and +1.14% compared to deep learning). The proposed application has a vocabulary that covers over ten thousand plant species from Europe and adjacent countries and provides a powerful methodology for improving biodiversity mapping, restoration, and conservation biology.

Overproduction of reactive oxygen species and antioxidant superoxide dismutases (SOD1, SOD2) dysregulation contribute to chronic inflammation such as generated in inflammatory bowel diseases (IBD). A kinetic redox shotgun proteomic strategy (OcSILAC for Oxidized cysteine Stable Isotope Labelling by Amino acids in Cell culture) was used to explore the lipopolysaccharide (LPS) effects including LPS‐induced oxidation and inflammation cascades on a dedicated intestinal epithelial cell line (HT29‐MD2) together with the potential mitigating role of a Mn‐based SOD‐mimic Mn1. While LPS induced transient oxidative damages at early times (15 min), cells incubated with Mn1 showed, in this time frame, a significantly reduced cysteine oxidation, highlighting Mn1 antioxidant properties. Over time, cysteine oxidation of LPS‐treated cells was counteracted by an overexpression of antioxidant proteins (SOD1, NQO1) and a late (6 h) preponderant increase in SOD2 level. Mn1, when co‐incubated with LPS, attenuated the level of most LPS‐modified proteins, that is, proteins involved in the inflammatory response. Our results highlight Mn1 as a potentially effective antioxidant and anti‐inflammatory agent to consider in the treatment of IBD, as well as a useful tool for exploring the interconnection between oxidative stress and inflammation.

RNAs are a vast reservoir of untapped drug targets. Structure-based virtual screening (VS) identifies candidate molecules by leveraging binding site information, traditionally using molecular docking simulations. However, docking struggles to scale with large compound libraries and RNA targets. Machine learning offers a solution but remains underdeveloped for RNA due to limited data and practical evaluations. We introduce a data-driven VS pipeline tailored for RNA, utilizing coarse-grained 3D modeling, synthetic data augmentation, and RNA-specific self-supervision. Our model achieves a 10,000x speedup over docking while ranking active compounds in the top 2.8% on structurally distinct test sets. It is robust to binding site variations and successfully screens unseen RNA riboswitches in a 20,000-compound in-vitro microarray, with a mean enrichment factor of 2.93 at 1%. This marks the first experimentally validated success of structure-based deep learning for RNA VS.

Metal Binder Jetting (MBJ) is an indirect, non-melting additive manufacturing (AM) technique. It involves selectively spraying a polymer binder onto a powder bed, layer by layer. A green part is obtained after curing and depowdering. A debinding stage removes polymer residues before final sintering to perfect mechanical properties. Poorly controlled debinding leads to carbon contamination, causing carbide precipitation during sintering and impacting the microstructure. Thermogravimetric experiments combined with SEM observations allowed to better understand the thermal and thermochemical degradation of the polymer to achieve optimal sintering. A new debinding cycle has been developed to completely remove the polymer while limiting oxidation of the surface of the powder particles.

The abundance of literature on institutional change in management studies poses a vexing problem: how to address the complex causal relations that institutional change entails? In this paper, we tackle that question by investigating through configurational analysis on the causal relations between the individual, organizational, and field-level drivers of institutional change. For that purpose, we develop an integrative framework for the study of institutional change and apply that framework in order to identify the configurations of conditions that lead to diverge from or converge with the institutionalized patterns of organizing in the field of hospitals. Our Qualitative Comparative Analysis of 115 cases of projects that were aimed at implementing a divergent template for organizing in French hospitals – namely: the Outpatient Shift – reveals four configurations of conditions that propelled that institutional change from 2015 to 2017, and three configurations of conditions that reinforced preexisting patterns of organizing during that period of time. Our work advances research on the drivers of institutional change in three ways. Firstly, our results contribute to bridging lasting gaps in the theory of institutional change, by demonstrating that institutional change results from interaction patterns between the different types of drivers that this theory has successively emphasized. Secondly, we provide evidence-based guidance for designing and implementing successful change strategies that integrate those drivers into the organizational context in which they take place. Thirdly, we propose a replicable method for exploring the new research avenues that our configurational perspective opens up for future studies on institutional change.

Background The ataxia-telangiectasia mutated (ATM) kinase phosphorylates and activates several downstream targets that are essential for DNA damage repair, cell cycle inhibition and apoptosis. Germline biallelic inactivation of the ATM gene causes ataxia-telangiectasia (A-T), and heterozygous pathogenic variant (PV) carriers are at increased risk of cancer, notably breast cancer. This study aimed to investigate whether DNA methylation profiling can be useful as a biomarker to identify tumors arising in ATM PV carriers, which may help for the management and optimal tailoring of therapies of these patients. Methods Breast tumor enriched DNA was prepared from 2 A-T patients, 27 patients carrying an ATM PV, 6 patients carrying a variant of uncertain clinical significance and 484 noncarriers enrolled in epidemiological studies conducted in France and Australia to investigate genetic and nongenetic factors involved in breast cancer susceptibility. Genome-wide DNA methylation analysis was performed using the Illumina Infinium HumanMethylation EPIC and 450K BeadChips. Correlation between promoter methylation and gene expression was assessed for 10 tumors for which transcriptomic data were available. Results We found that the ATM promoter was hypermethylated in 62% of tumors of heterozygous PV carriers compared to the mean methylation level of ATM promoter in tumors of noncarriers. Gene set enrichment analyses identified 47 biological pathways enriched in hypermethylated genes involved in neoplastic, neurodegenerative and metabolic-related pathways in tumor of PV carriers. Among the 327 differentially methylated promoters, promoters of ARHGAP40, SCGB3A1 (HIN-1), and CYBRD1 (DCYTB) were hypermethylated and associated with a lower gene expression in these tumors. Moreover, using three different deep learning algorithms (logistic regression, random forest and XGBoost), we identified a set of 27 additional biomarkers predictive of ATM status, which could be used in the future to provide evidence for or against pathogenicity in ATM variant classification strategies. Conclusions We showed that breast tumors that arise in women who carry an ATM PV display a specific genome-wide DNA methylation profile. Specifically, the methylation pattern of 27 key gene promoters was predictive of ATM PV status of the women. These genes may also represent new medical prevention and therapeutic targets for these women.

The increased frequency of rainfall‐triggered geohazards has led to more disruptions of transportation networks in recent years. This study proposes an integrated framework for resilience assessment of transportation networks, where the highway disruption scenarios are simulated using a traffic model and a developed geohazard threat model based on real‐world datasets. In the framework, we provide a resilience‐oriented indicator that integrates traffic flow and geohazard threat, upon which to identify critical elements of the highway network and the geohazard‐prone sites. To enhance the performance of the highway network under disruptions, we design multiple strategies including reinforcing highway segments and preventing potential geohazards. We apply the proposed approach to a realistic case study of the highway network of Shaanxi province in China. The results of the analysis demonstrate the validity and reliability of the resilience‐oriented indicator, and illustrate that a mixture of reinforcement strategies provides better improvement in system performance under scenarios with different traffic demand level.

Most transmission system operators (TSOs) currently use seasonally steady‐state models considering limiting weather conditions that serve as reference to compute the transmission capacity of overhead power lines. The use of dynamic line rating (DLR) models can avoid the construction of new lines, market splitting, false congestions, and the degradation of lines in a cost‐effective way. DLR can also be used in the long run in grid extension and new power capacity planning. In the short run, it should be used to help operate power systems with congested lines. The operation of the power systems is planned to have the market trading into account; thus, it computes transactions hours ahead of real‐time operation, using power flow forecasts affected by large errors. In the near future, within a “smart grid” environment, in real‐time operation conditions, TSOs should be able to rapidly compute the capacity rating of overhead lines using DLR models and the most reliable weather information, forecasts, and line measurements, avoiding the current steady‐state approach that, in many circumstances, assumes ampacities above the thermal limits of the lines. This work presents a review of the line rating methodologies in several European countries and the United States. Furthermore, it presents the results of pilot projects and studies considering the application of DLR in overhead power lines, obtaining significant reductions in the congestion of internal networks and cross‐border transmission lines.

The European Paediatric Soft Tissue Sarcoma Study Group (EpSSG) RMS 2005 trial evaluated maintenance chemotherapy in high-risk rhabdomyosarcoma (RMS). Patients were randomly assigned to either discontinue treatment (standard arm) or receive six 28-day cycles of vinorelbine (25 mg/m ² ) once per day on days 1, 8, and 15, plus once daily low-dose cyclophosphamide (25 mg/m ² ; experimental arm). Initial results showed improved overall survival (OS), but disease-free survival (DFS) improvement was not statistically significant. This report presents mature survival outcomes after extended follow-up. Between April 2006 and December 2016, 186 patients were enrolled in the standard arm and 185 in the experimental arm. After a median follow-up of 122.1 months from diagnosis and 114 months from random assignment, recurrence, progression, or death occurred in 103 patients (61 standard arm, 42 experimental arm). The 10-year DFS was 66.5% (95% CI, 59 to 74) in the standard arm versus 77.1% (95% CI, 70.3 to 82.5) in the experimental arm ( P = .025). Corresponding 10-year OS rates were 70.8% (95% CI, 63.3 to 77.0) and 82.9% (95% CI, 76.6 to 87.7; P = .0099). Long-term results of the RMS2005 trial confirm the survival benefit of maintenance chemotherapy with vinorelbine and low-dose cyclophosphamide for patients with high-risk RMS.

Background The epidemiology of childhood cancer in Afro-descendant (AD) populations is poorly described. We performed a descriptive study of the distribution, incidence, and survival of children with cancer in the French West Indies (FWI) and French Guiana (FG). Methods We included all patients aged 0–17 diagnosed with cancer or benign intracranial tumor between January 2011 and December 2021 and living in the FWI/FG area at time of diagnosis. The cases were identified from the French national registry of childhood cancer and cross-referenced with local sources. Incidence rates were calculated, and compared to that of mainland France by standardized incidence ratios (SIR). Vital status was completed up to the 31st of December 2023 (date of point). Relapses were identified and documented in pediatric reference centers in mainland France and local centers. The 5-year overall survival (5yOS) and event-free survival (5yEFS) were estimated using Kaplan–Meier method. Findings We identified 368 patients (26% leukemias, 21% central nervous system tumors, 12% lymphomas, and 41% others). The average age at diagnosis was 8.8 years (Range: 0.1–17.8), with 52% boys. The median follow-up was 4.4 years (Range: 0.1–12.3). The age standardized rates for all cancers was lower than in mainland France (124.9 vs 162.6 per million-year for children under 18 years old, SIR = 0.77 [95% CI: 0.69–0.85]). The 5yOS was 78.9% [95% CI: 73.9–83.0] and 5yEFS was 69.3% [95% CI: 63.9–74.0]. The 5yOS for the 0–14 age group was 81.2% [95% CI: 76.9–85.5]. Interpretation This first registry-based study of childhood cancer in the FWI and FG shows that our patients with childhood cancer, treated in a country with a high standard of health care, has resulted in overall survival comparable to that of European and North American children. Funding The authors received no financial support.

Galactic Supernova remnants (SNRs) are likely to be significant sources of cosmic rays (CRs) up to the knee of the CR spectrum. They produce gamma rays in the very-high-energy (VHE) range. About a dozen SNRs emitting VHE gamma rays have been detected by current instruments and it is expected that many more will be detected by future instruments. However, the details of particle acceleration at SNRs, and the mechanisms producing VHE gamma rays remain poorly understood. We studied the population of SNRs in the TeV range and its properties by confronting simulated samples to the catalogue of VHE gamma-ray sources from the H.E.S.S. Galactic Plane Survey (HGPS) under consideration of the multi-dimensional detection threshold of the HGPS. This allows us to address fundamental questions concerning particle acceleration at SNR shocks. Particularly, what is the efficiency of particle acceleration? What is the spectrum of accelerated particles? Is the VHE gamma-ray emission dominated by hadronic or leptonic interactions? We present here the first systematic exploration of the SNR-population parameter space relevant to our model. We identify preferred parameter combinations for which ≳ 90% of the Monte Carlo realisations are in agreement with VHE gamma-ray data and exclude parts of the parameter space in contradiction with the HGPS data. One finding is a preference for large hadron domination (lower electron-to-proton fractions of Kep ≲ 10−4.5) in the simulated SNRs, but despite this a significant fraction (∼ 50%) of the detectable simulated SNRs are dominated by leptonic emission.

Accurately predicting the remaining useful life (RUL) of aircraft engines across varying working conditions is challenging, particularly due to the absence of labeled data in the target domain. Traditional approaches in unsupervised domain adaptation (UDA) have largely focused on working-condition-invariant (domain-invariant) features, often overlooking the valuable contributions that working-condition-related (domain-related) features can provide. To address this limitation, we propose a novel UDA framework that explores the potential contributions of working-condition-related features and leverages the synergy between working-condition-invariant and working-condition-related features for cross-domain RUL prediction. Our approach involves the development of two specialized feature generators, one for working-condition-invariant features and another for working-condition-related features. We employ orthogonality constraints to optimize the expression of both feature types and ensure their independence within the feature space, which effectively reduces interference between them. Combined with a domain discriminator and a domain classifier, this setup allows our RUL predictor to harness the combined strengths of both feature types under varying working conditions. The effectiveness of our approach is validated through extensive experiments across twelve distinct working-condition scenarios, demonstrating a significant improvement in prediction accuracy, with a 9.3% reduction in root-mean-square error (RMSE).

Unlike developmental biologists, paleoanthropologists primarily investigate development using skeletal remains, specifically fossilized and already‐formed bones and teeth. Focusing on peri‐ and/or postnatal growth, they reconstruct development from fragmented “snapshots” of individual trajectories at various ontogenetic stages. These constraints prompt a discussion of what defines development versus growth, and its boundaries in studies of hominin evolution. We explore how paleoanthropologists address the limitations of the fossil record by using diverse methodological and theoretical frameworks to identify developmental markers despite missing data. Finally, we discuss the potential of the “Extended Evolutionary Synthesis,” which calls for a greater focus on developmental processes in interpreting phenotypic variation in the fossil record.

The challenges of drug discovery from hit identification to clinical development sometimes involves addressing scaffold hopping issues, in order to optimise molecular biological activity or ADME properties, or mitigate toxicology concerns of a drug candidate. Docking is usually viewed as the method of choice for identification of isofunctional molecules, i. e. highly dissimilar molecules that share common binding modes with a protein target. However, the structure of the protein may not be suitable for docking because of a low resolution, or may even be unknown. This problem is frequently encountered in the case of membrane proteins, although they constitute an important category of the druggable proteome. In such cases, ligand‐based approaches offer promise but are often inadequate to handle large‐step scaffold hopping, because they usually rely on molecular structure. Therefore, we propose the Interaction Fingerprints Profile (IFPP), a molecular representation that captures molecules binding modes based on docking experiments against a panel of diverse high‐quality proteins structures. Evaluation on the LH benchmark demonstrates the interest of IFPP for identification of isofunctional molecules. Nevertheless, computation of IFPPs is expensive, which limits its scalability for screening very large molecular libraries. We propose to overcome this limitation by leveraging Metric Learning approaches, allowing fast estimation of molecules IFPP similarities, thus providing an efficient pre‐screening strategy that in applicable to very large molecular libraries. Overall, our results suggest that IFPP provides an interesting and complementary tool alongside existing methods, in order to address challenging scaffold hopping problems effectively in drug discovery.