7 research outputs found
From Big Data to Prediction: Machine Learning for source attribution in a One Health approach - A case study on L. monocytogenes
No full textInternational audienceSource attribution is essential for food safety and public health, enabling the identification of contamination origins and supporting outbreak response. Listeria monocytogenes (Lm), a widespread foodborne pathogen, causes listeriosis, a severe illness with high fatality rates among vulnerable populations. Machine learning (ML) has shown promise in addressing source attribution, but prior studies often relied on small, temporally and geographically limited datasets.In this study, we leveraged a high-quality dataset of 5,366 Lm isolates collected between 1997 and 2024 from six food sources (mammals’ meat, birds’ meat, dairy, fish and seafood, vegetables, and fruits) across five continents5. We evaluated four genomic feature sets: core genome multilocus sequence typing (cgMLST), pangenome gene presence/absence, genetic variants, and 21-mers. LightGBM model was trained with a stratified 80/20 train-test split, ensuring balanced representation of source, location, and date. Hyperparameters were optimized using randomized search with 10-fold cross-validation.Tuned models outperformed default configurations, improving F1-scores by 1–2% and achieving scores between 72–74%, with cgMLST yielding the highest performance. Model accuracy varied across food sources; for example, fruit-derived isolates consistently achieved F1-scores above 80%, especially with cgMLST and pangenome-based features. While 16% of isolates were correctly classified by only one feature type, 9% were misclassified across all, and combining feature sets did not reduce overall error rates.To address misclassification, we restructured the source categories to reflect practical groupings used in the food industry. Grouping meats together and separating plant-based and animal-derived products led to improved model performance, with F1-scores reaching 78%.We further compared LightGBM with Random Forest, a widely used model in the literature. While overall performance was similar, error distributions differed between the two. Finally, we assessed model robustness across time and geography. Accuracy declined when predicting isolates collected years after training or from different continents, indicating temporal and spatial limitations.Overall, our findings confirm the potential of ML-based source attribution for Lm on a global scale. Our results suggest that strains can be difficult to discriminate when evolved in same production facility. We also highlight the importance of diverse, representative datasets for improving predictive performance and reproducibility among time and space
Nettoyer pour Innover : Améliorer la qualité des données publiques de pathogènes alimentaires pour les utiliser dans des modèles d’IA
No full textNational audienceEn sécurité alimentaire, l’Intelligence Artificielle (IA) a récemment été utilisée afin de prédire l’origine, la virulence ou la résistance d’agents pathogènes (AP). Cependant, ces modèles ont été entrainés sur des jeux de données restreints [1-3]. A l’ère du Big Data, l’étude des AP bénéficie de vastes informations publiques grâce notamment aux réseaux de surveillance et aux projets de recherche. La démocratisation du séquençage a entraîné une croissance exponentielle du nombre de génomes bactériens disponibles dans les bases de données publiques. Malgré des tentatives d’harmonisation, les métadonnées contextuelles de ces souches restent souvent hétérogènes et non structurées du fait de leur provenance variée. Ainsi, l’application de l’IA aux données massivement disponibles, grâce aux politiques d’Open Data, nécessite le développement d’outils et de méthodes d’harmonisation de ces métadonnées [4,5]. L’objectif de cette étude a été de développer des méthodes automatisées de normalisation et de standardisation permettant l’interopérabilité et l’exploitation de ces métadonnées et données génomiques publiques par la communauté scientifique. Pour cela, nous nous sommes principalement appuyés sur deux référentiels internationaux : Geopy pour les données géographiques et l’ontologie Foodex2 pour décrire et hiérarchiser les sources alimentaires. Ces métadonnées ont été ensuite croisées avec les données d’annotation génomique (gènes d’intérêt, SNP, cgMLST...).Nous avons collecté deux bases de données d’AP impliqués dans des toxi-infections alimentaires à partir de NCBI. Seules les souches présentant une origine clairement définie ont été sélectionnés pour la suite : 46.977 souches de Listeria monocytogenes (Lm) et 10.995 souches de Vibrio parahaemolyticus (Vp). L’application des méthodes décrites a permis de réduire la diversité des métadonnées géographiques d’un facteur 3X et de l’origine des souches d’un facteur 4X. Les bases obtenues présentent ainsi des souches provenant de 69 (Lm) et 41 (Vp) pays différents collectées entre 1900 et 2024 et provenant de 368 (Lm) et 77 (Vp) origines différentes (aliments, environnements ou humains). Ces étapes de nettoyage sont cruciales pour l’utilisation des modèles d'IA dédiés à la compréhension des épidémies et l’anticipation des risques construits à partir de données génomiques. Elles pourront être appliquées à d’autres AP d’intérêts en sécurité alimentaire, en santé humaine et animale
From Big Data to Prediction: Machine Learning for source attribution in a One Health approach - A case study on L. monocytogenes
No full textInternational audienceSource attribution is essential for food safety and public health, enabling the identification of contamination origins and supporting outbreak response. Listeria monocytogenes (Lm), a widespread foodborne pathogen, causes listeriosis, a severe illness with high fatality rates among vulnerable populations. Machine learning (ML) has shown promise in addressing source attribution, but prior studies often relied on small, temporally and geographically limited datasets.In this study, we leveraged a high-quality dataset of 5,366 Lm isolates collected between 1997 and 2024 from six food sources (mammals’ meat, birds’ meat, dairy, fish and seafood, vegetables, and fruits) across five continents5. We evaluated four genomic feature sets: core genome multilocus sequence typing (cgMLST), pangenome gene presence/absence, genetic variants, and 21-mers. LightGBM model was trained with a stratified 80/20 train-test split, ensuring balanced representation of source, location, and date. Hyperparameters were optimized using randomized search with 10-fold cross-validation.Tuned models outperformed default configurations, improving F1-scores by 1–2% and achieving scores between 72–74%, with cgMLST yielding the highest performance. Model accuracy varied across food sources; for example, fruit-derived isolates consistently achieved F1-scores above 80%, especially with cgMLST and pangenome-based features. While 16% of isolates were correctly classified by only one feature type, 9% were misclassified across all, and combining feature sets did not reduce overall error rates.To address misclassification, we restructured the source categories to reflect practical groupings used in the food industry. Grouping meats together and separating plant-based and animal-derived products led to improved model performance, with F1-scores reaching 78%.We further compared LightGBM with Random Forest, a widely used model in the literature. While overall performance was similar, error distributions differed between the two. Finally, we assessed model robustness across time and geography. Accuracy declined when predicting isolates collected years after training or from different continents, indicating temporal and spatial limitations.Overall, our findings confirm the potential of ML-based source attribution for Lm on a global scale. Our results suggest that strains can be difficult to discriminate when evolved in same production facility. We also highlight the importance of diverse, representative datasets for improving predictive performance and reproducibility among time and space
Proteogenomic reconstruction of organ-specific metabolic networks in an environmental sentinel species, the amphipod Gammarus fossarum
No full textInternational audienceMetabolic pathways are affected by the impacts of environmental contaminants underlying a large variability of toxic effects across different species. However, the systematic reconstruction of metabolic pathways remains limited in environmental sentinel species due to the lack of available genomic data in many taxa of animal diversity. In this study we used a multi-omics approach to reconstruct the most comprehensive map of metabolic pathways for a crustacean model in biomonitoring, the amphipod Gammarus fossarum in order to improve the knowledge of the metabolism of this sentinel species.We revisited the assembly of RNA-seq data by de novo approaches to reduce RNA contaminants and transcript redundancy. We also acquired extensive mass spectrometry shotgun proteomic data on several organs from a reference population of G. fossarum males and females to identify organ-specific metabolic profiles. The G. fossarum metabolic pathway reconstruction (available through the metabolic database GamfoCyc) was performed by adapting the genomic tool CycADS and we identified 377 pathways representing 7630 annotated enzymes, 2610 enzymatic reactions and the expression of 858 enzymes was experimentally validated by proteomics. To our knowledge, our analysis provides for the first time a systematic metabolic pathway reconstruction and the proteome profiles of these pathways at the organ level in this sentinel species. As an example, we show an elevated abundance in enzymes involved in ATP biosynthesis and fatty acid beta-oxidation indicative of the high-energy requirement of the gills, or the key anabolic and detoxification role of the hepatopancreatic caeca, as exemplified by the specific expression of the retinoid biosynthetic pathways and glutathione synthesis.In conclusion, the multi-omics data integration performed in this study provides new resources to investigate metabolic processes in crustacean amphipods and their role in mediating the effects of environmental contaminant exposures in sentinel species. Synopsis: This study provide the first evidence that it is possible to combine multiple omics data to exhaustively describe the metabolic network of a model species in ecotoxicology, Gammarus fossarum, for which a reference genome is not yet available
Eyewitness accounts of 'the Indies' in the Later Medieval West: reading, reception, and re-use (c. 1300-1500)
No full textDespite increased mercantile and missionary contact between the Latin West and India and China between the thirteenth and fifteenth centuries, scholars have often
noted that Western Europe's knowledge of India, as judged by geographical texts from the period, changed surprisingly little during this time. This thesis employs
some of the methodologies of reception studies in order to investigate the role played by first-hand travel accounts in the construction and change of concepts of the Indies
during the fourteenth and fifteenth centuries. It investigates in particular the reception in Italy, France and England of the information about the area known as
India or the 'three Indies' presented in the texts produced by two Italian travellers to the East: the Divisament dou monde of the Venetian merchant Marco Polo (c. 1298),
and the Relatio of the Franciscan missionary Odorico da Pordenone (1330).
The thesis falls into three distinct parts. In the first section, I contextualise the project with a broad survey of the Latin European ideas of India in the late thirteenth
and early fourteenth centuries and with an outline of the travellers' journeys and their contexts. The second part of the thesis provides a broad overview of the
circumstances of diffusion of the two travel accounts in England, France and Italy over the fourteenth and fifteenth centuries, before conducting a detailed, manuscriptbased
investigation of the ways in which the two accounts of India were approached by their early readers. This investigation focuses principally upon the presentation
and possible modes of reception of the texts' geographical and ethnographic details and relies heavily on the evidence of presentation, paratext and the traces of reading
present in the physical texts of the accounts.
The third and final part of the thesis considers the evidence of the reception of elements from first-hand travel accounts in other textual and cartographic
productions. Proceeding on the basis of case studies, it demonstrates that first-hand accounts of 'the Indies' were used by the authors and compilers of cosmo graphical
texts in this period in a variety of ways. It suggests, however, that the manner and context of the deployment of elements from such accounts often tended to assimilate
these with, rather than distinguish them from, the writings of accepted authorities. This section also contrasts the way that details from travel accounts were re-used in
texts with the way the same information was handled in the composition of maps. Finally, by analysis of the ways eyewitness accounts of the Indies were re-used in
certain ambiguous and comic texts produced in this period, the thesis sheds light on an underexplored aspect of the reception both of eyewitness information and of the
genres in which it appeared. The appendices contain tables presenting information relative to the manuscripts discussed that support the arguments presented in section
two
Global, regional, and national under-5 mortality, adult mortality, age-specific mortality, and life expectancy, 1970-2016: a systematic analysis for the Global Burden of Disease Study 2016
No full textBackground Detailed assessments of mortality patterns, particularly age-specific mortality, represent a crucial input that enables health systems to target interventions to specific populations. Understanding how all-cause mortality has changed with respect to development status can identify exemplars for best practice. To accomplish this, the Global Burden of Diseases, Injuries, and Risk Factors Study 2016 (GBD 2016) estimated age-specific and sex-specific all-cause mortality between 1970 and 2016 for 195 countries and territories and at the subnational level for the five countries with a population greater than 200 million in 2016.Methods We have evaluated how well civil registration systems captured deaths using a set of demographic methods called death distribution methods for adults and from consideration of survey and census data for children younger than 5 years. We generated an overall assessment of completeness of registration of deaths by dividing registered deaths in each location-year by our estimate of all-age deaths generated from our overall estimation process. For 163 locations, including subnational units in countries with a population greater than 200 million with complete vital registration (VR) systems, our estimates were largely driven by the observed data, with corrections for small fluctuations in numbers and estimation for recent years where there were lags in data reporting (lags were variable by location, generally between 1 year and 6 years). For other locations, we took advantage of different data sources available to measure under-5 mortality rates (U5MR) using complete birth histories, summary birth histories, and incomplete VR with adjustments; we measured adult mortality rate (the probability of death in individuals aged 15-60 years) using adjusted incomplete VR, sibling histories, and household death recall. We used the U5MR and adult mortality rate, together with crude death rate due to HIV in the GBD model life table system, to estimate age-specific and sex-specific death rates for each location-year. Using various international databases, we identified fatal discontinuities, which we defined as increases in the death rate of more than one death per million, resulting from conflict and terrorism, natural disasters, major transport or technological accidents, and a subset of epidemic infectious diseases; these were added to estimates in the relevant years. In 47 countries with an identified peak adult prevalence for HIV/AIDS of more than 0.5% and where VR systems were less than 65% complete, we informed our estimates of age-sex-specific mortality using the Estimation and Projection Package (EPP)-Spectrum model fitted to national HIV/AIDS prevalence surveys and antenatal clinic serosurveillance systems. We estimated stillbirths, early neonatal, late neonatal, and childhood mortality using both survey and VR data in spatiotemporal Gaussian process regression models. We estimated abridged life tables for all location-years using age-specific death rates. We grouped locations into development quintiles based on the Sociodemographic Index (SDI) and analysed mortality trends by quintile. Using spline regression, we estimated the expected mortality rate for each age-sex group as a function of SDI. We identified countries with higher life expectancy than expected by comparing observed life expectancy to anticipated life expectancy on the basis of development status alone.Findings Completeness in the registration of deaths increased from 28% in 1970 to a peak of 45% in 2013; completeness was lower after 2013 because of lags in reporting. Total deaths in children younger than 5 years decreased from 1970 to 2016, and slower decreases occurred at ages 5-24 years. By contrast, numbers of adult deaths increased in each 5-year age bracket above the age of 25 years. The distribution of annualised rates of change in age-specific mortality rate differed over the period 2000 to 2016 compared with earlier decades: increasing annualised rates of change were less frequent, although rising annualised rates of change still occurred in some locations, particularly for adolescent and younger adult age groups. Rates of stillbirths and under-5 mortality both decreased globally from 1970. Evidence for global convergence of death rates was mixed; although the absolute difference between age-standardised death rates narrowed between countries at the lowest and highest levels of SDI, the ratio of these death rates-a measure of relative inequality-increased slightly. There was a strong shift between 1970 and 2016 toward higher life expectancy, most noticeably at higher levels of SDI. Among countries with populations greater than 1 million in 2016, life expectancy at birth was highest for women in Japan, at 86.9 years (95% UI 86.7-87.2), and for men in Singapore, at 81.3 years (78.8-83.7) in 2016. Male life expectancy was generally lower than female life expectancy between 1970 and 2016, and the gap between male and female life expectancy increased with progression to higher levels of SDI. Some countries with exceptional health performance in 1990 in terms of the difference in observed to expected life expectancy at birth had slower progress on the same measure in 2016.Interpretation Globally, mortality rates have decreased across all age groups over the past five decades, with the largest improvements occurring among children younger than 5 years. However, at the national level, considerable heterogeneity remains in terms of both level and rate of changes in age-specific mortality; increases in mortality for certain age groups occurred in some locations. We found evidence that the absolute gap between countries in age-specific death rates has declined, although the relative gap for some age-sex groups increased. Countries that now lead in terms of having higher observed life expectancy than that expected on the basis of development alone, or locations that have either increased this advantage or rapidly decreased the deficit from expected levels, could provide insight into the means to accelerate progress in nations where progress has stalled. Copyright (C) The Author(s). Published by Elsevier Ltd. This is an Open Access article under the CC BY 4.0 license
Directorio de consultores, recursos y sitios de Internet relacionados con bibliotecas mexicanas = Directory of consultants, resources & Internet sites relating to Mexican libraries
No full textThis guidebook was compiled for: 1) foreign librarians, 2) Mexican students enrolled in programs of study in the field of librarianship or in certificate programs in library science, 3) volunteers or those in practicum service, 4) paraprofessionals or the recently degreed Mexican librarian, 5) non-experts. This work, a bilingual annotated directory, contains basic information on a wide range of resources relevant to librarianship as presently practiced in Mexico: books, articles, useful web pages, events, possible contacts in institutions. To find specific phrases or words use your navigator’s BUSCAR/FIND search tool, or scroll down. There is no intention to publish the list at this time. The information is being provided as a free service. This directory database is not exhaustive; the user is encouraged to verify all data from the source. Please provide us with your opinion concerning this Directory. All additions, suggestions, or modifications will be welcome. To contact the compiler, email: William Abrams Indexing Services, [email protected] . Your comments will help us to improve future editions. Terms of Use: This Directory is not copyrighted. It is a document in the public domain. No rights are reserved, either for the original or for derivative works. The file may be freely copied without prior permission, preferably using a CD-ROM data disc (but if access is from the website, one should first verify that the download has completed before copying). (Abstract taken verbatim from author's)
