43569 research outputs found
Sort by
SNPeBoT: a tool for predicting transcription factor allele specific binding
No full textBackground: Mutations in non-coding regulatory regions of DNA may lead to disease through the disruption of transcription factor binding. However, our understanding of binding patterns of transcription factors and the effects that changes to their binding sites have on their action remains limited. To address this issue we trained a Deep learning model to predict the effects of Single Nucleotide Polymorphisms (SNP) on transcription factor binding. Allele specific binding (ASB) data from Chromatin Immunoprecipitation sequencing (ChIP-seq) experiments were paired with high sequence-identity DNA binding Domains assessed in Protein Binding Microarray (PBM) experiments. For each transcription factor a paired DNA binding Domain was selected from which we derived E-score profiles for reference and alternate DNA sequences of ASB events. A Convolutional Neural Network (CNN) was trained to predict whether these profiles were indicative of ASB gain/loss or no change in binding. 18211 E-score profiles from 113 transcription factors were split into train, validation and test data. We compared the performance of the trained model with other available platforms for predicting the effect of SNP on transcription factor binding. Our model demonstrated increased accuracy and ASB recall in comparison to the best scoring benchmark tools. Conclusion: In this paper we present our model SNPeBoT (Single Nucleotide Polymorphism effect on Binding of Transcription Factors) in its standalone and web server form. The increased recovery and prediction accuracy of allele specific binding events could prove useful in discovering non-coding mutations relevant to disease.The work was supported by grants PID2020-113203RB-I00, PID2023-150068OB-I00 and “Unidad de Excelencia María de Maeztu” (ref: CEX2018-000792-M), funded by the MCIN and the AEI https://doi.org/10.13039/501100011033, MCIUN/AEI/10.13039/501100011033/FEDER, UE as well as an FPU scholarship (ref: FPU22/02303) and an SGR from the Generalitat de Catalunya (ref: 4413015318- J.SELENT/SGR-22)
Early-life growth and emotional, behavior and cognitive outcomes in childhood and adolescence in the EU child cohort network: individual participant data meta-analysis of over 109,000 individuals
No full textBackground: Fetal and infant development might be critical for cognitive outcomes and psychopathology later in life. We assessed the associations of birth characteristics and early life growth with behavior and cognitive outcomes from childhood to adolescence. Methods: We used harmonized data of 109,481 children from 8 European birth cohorts. Birth weight, gestational age, and body mass index (BMI) tertiles at the age of 2 years were used as the exposure variables. Outcomes included internalizing and externalizing problems and attention-deficit hyperactivity disorder (ADHD), autism spectrum disorder (ASD), and non-verbal intelligence quotient (Non-verbal IQ) in childhood (4-10 years), early adolescence (11-16 years), and late adolescence (17-20 years). We used 1-stage individual participant data meta-analyses using generalized linear models. Findings: A one-week older gestational age was associated with lower scores for internalizing problems (difference -0·48 (95% CI: -0·59, -0·37)), externalizing problems (difference -0·34 (95% CI: -0·44, -0·23)), and ADHD symptoms (difference -0·38 (95% CI: -0·49, -0·27)), and with higher scores for non-verbal IQ (difference 0·65 (95% CI: 0·41, 0·89)). As compared to term birth, preterm birth was associated with higher internalizing problems (difference 3·43 (95% CI: 2·52, 4·33)) and externalizing problems (difference 2·31 (95% CI: 1·16, 3·46)), ADHD symptoms (difference 4·15 (95% CI: 3·15, 5·16)), ASD symptoms (difference 3·23 (95% CI: 0·37, 6·08)), and lower non-verbal IQ (difference -5·44 (95% CI: -7·44, -3·44)). Small size for gestational age at birth (SGA) in comparison with appropriate size for gestational age (AGA) was associated with higher ADHD symptoms (difference 4·88 (95% CI: 3·87, 5·90)) and lower Non-verbal IQ (difference -7·02 (95% CI: -8·84, -5·21)). Large size for gestational age at birth was associated with lower ADHD symptoms (difference -1·09 (95% CI: -1·73, 0·45)) and higher non-verbal IQ (difference 2·47 (95% CI: 0·77, 4·18)). Explorative analyses showed that as compared to children with an appropriate size for gestational age at birth and a normal BMI at the age of 2 years, children born SGA who remained small at 2 years had the lowest non-verbal IQ score (difference -8·14 percentiles (95% CI: -11·89, -4·39)). Interpretation: Both fetal and early childhood growth are associated with emotional, behavioral and cognitive outcomes throughout childhood and adolescence. Compensatory infant growth might partly attenuate the adverse effects of suboptimal fetal growth. Future studies are needed to identify the potential for optimizing mental health outcomes in new generations by improving early-life growth. Funding: This project received funding from the European Union's Horizon 2020 research and innovation programme (LIFECYCLE, grant agreement No 733206, 2016; EUCAN-Connect grant agreement No 824989; ATHLETE, grant agreement No 874583).The authors are grateful to researchers and participants from the EU Child Cohort Network, ATHLETE, and LifeCycle who have supported and contributed to each cohort included in this study. In addition, acknowledgments are sent to the DataSHIELD and Molgenis team. Cohort specific acknowledgements can be found in Supplementary Text S5. This project received funding from the European Union's Horizon 2020 research and innovation programme (LIFECYCLE, grant agreement No 733206, 2016; EUCAN-Connect grant agreement No 824989; ATHLETE, grant agreement No 874583). Cohort specific funding details can be found in Supplementary Text S6
Deep research agentic framework for mitigating bias in AI-driven healthcare diagnostics
No full textTreball fi de màster de: Erasmus Mundus joint Master in Artificial Intelligence (EMAI)Supervisors: Vicenç Gómez & Leo Anthony Celi Academic
Tutor: Vicenç GómezTransformer-scale language models can now ace many medical exams, but their frozen parametric memory risks propagating outdated guidelines and systemic bias to the bedside. To counter this, we re-imagine the diagnostic assistant as a navigator that plans, retrieves, executes code, and verifies evidence rather than guessing from memory. We introduce DeepMed, a 4 B-parameter multi-agent framework who attempts to switch the paradigm of medical assistances from diagnostic oracles to information retrievers. Agents invoke external tools via the open Model Context Protocol (MCP), including M3, a natural-language gateway to the MIMIC-IV EHR, and a sandboxed Python REPL for on-the-fly calculations. Performance is audited on the newly proposed MedBrowseComp benchmark (1 089 quarterly-regenerating, multi-hop oncology related queries), legacy QA suites, the EquityMedQA counter factual set, and the EHRSQL challenge. With just a 4 billion parameter LLM as the cognitive engine DeepMed achieves 26 %single-pass accuracy on MedBrowseComp, outperforming larger entreprise grade systems that rely on 10 to 100 times larger fine tuned models while running locally on a consumer laptop. On EquityMedQA it increases correctness from 50.8 % to 57.4%, a 13% relative reduction in demographic disparity. Coupling MCP to the M3 EHR interface lifts pass@1 on EHRSQL from 2% to 9%. By fusing agentic planning, typed tool use, and evidence-first reporting, DeepMed shows that bias-aware, verifiable clinical AI can be achieved without frontier-scale models or costly GPU clusters. The open-sourced multiagent framework, MCP server tool contributions like M3 and MedBrowseComp benchmark provide a reproducible path toward transparent, low-cost decision support in safety-critical healthcare settings
Novel geometric representations and optimized frameworks for deep learning models of interatomic potentials
No full textAquesta tesi explora avenços en potencials de xarxes neuronals per a simulacions atomístiques, abordant el doble repte de la precisió i l'eficiència. Mitjançant noves representacions geomètriques i marcs optimitzats, té com a objectiu enllaçar la química computacional tradicional amb enfocaments moderns basats en IA. Les contribucions principals inclouen el desenvolupament de TensorNet, un model equivariant que explota representacions de tensors cartesians per a predir amb precisió energies i forces moleculars. A més, aquest treball estén TensorNet per a incorporar els estats de càrrega i espín, ampliant així la seva aplicabilitat. Les millores en el marc TorchMD-Net, incloent el suport per a CUDA graphs i les optimitzacions de PyTorch 2.0, augmenten significativament el rendiment computacional i la usabilitat per a la dinàmica molecular. Combinant arquitectures de models innovadores amb una infraestructura de programari eficient, aquesta investigació avança el camp dels potencials interatòmics d'aprenentatge automàtic.This thesis explores advancements in neural network potentials for atomistic simulations, addressing the dual challenges of accuracy and efficiency. Through novel geometric representations and optimized frameworks, it aims to bridge traditional computational chemistry and modern AI-based approaches. Key contributions include the development of TensorNet, an equivariant model exploiting Cartesian tensor representations to predict molecular energies and forces accurately. Additionally, this work extends TensorNet to incorporate charge and spin states, broadening its applicability. Enhancements to the TorchMD-Net framework, including support for CUDA graphs and PyTorch 2.0 optimizations significantly improve computational performance and usability for molecular dynamics. By combining innovative model architectures with efficient software infrastructure, this research advances the field of machine learning interatomic potentialsPrograma de Doctorat en Tecnologies de la Informació i les Comunicacion
Pel carril lent: l’slow journalism a Catalunya a través de l’anàlisi de sis mitjans de comunicació
No full textDirectora: Mònica Terribas SalaTreball de fi de grau en Periodisme. Curs 2024-2025Aquest Treball de Fi de Grau de Periodisme analitza la presència del periodisme lent en l’ecosistema de mitjans de comunicació català. Té Catalunya exponents clars d’aquesta mena de periodisme? Com és el periodisme lent en català? A partir d’una revisió de la literatura existent que configura la definició actual del corrent slow journalism, s’analitza la correspondència de les bases teòriques del model amb la seva presència a Catalunya. Per fer-ho, s’analitzen com a mostra sis mitjans propers a les perspectives del periodisme lent: Catorze, Crític, El Temps, La Directa, La Mira i Núvol. A través d’una anàlisi qualitativa i d’entrevistes en profunditat amb els responsables del mitjà, s’exploren els formats, les línies temàtiques, les estructures organitzatives i els valors que apropen els projectes al periodisme lent.This Journalism Final Thesis examines the presence of slow journalism within the Catalan media ecosystem. Are there clear examples of this kind of journalism in Catalonia? How is this model portrayed in Catalan? Based on a review of the existing literature that defines the current concept of slow journalism, the study analyses how the theoretical foundations of the model align with its actual presence in Catalonia. To do so, it examines six media outlets that reflect the principles of slow journalism: Catorze, Crític, El Temps, La Directa, La Mira and Núvol. Through qualitative analysis and in-depth interviews with key figures at each outlet, the research explores formats, thematic focuses, organizational structures, and values that bring these projects closer to the ideals of slow journalism
Recent developments in event study estimation
No full textTreball fi de màster de: Master's Degree in Economics and Finance. PhD Track Program. Curs 2024-2025Tutora: Valeria GargiuloThis review examines recent methodological advances in the estimation of causal effects using difference-in-differences (DiD) frameworks, particularly under staggered treatment adoption with heterogeneous effects. While two-way fixed effects (TWFE) models remain widely used, they can yield misleading results when treatment effects vary across units or over time. To address these concerns, the review surveys a set of alternative estimators developed to improve identification and interpretation in such settings. These include dynamic group-time estimators, imputation-based methods, and doubly robust strategies. The paper complements this theoretical overview with a series of simulation studies designed to evaluate the empirical performance of these methods under varying degrees of treatment effect heterogeneity and error structure complexity. The findings highlight both the potential and the limitations of modern DiD estimators, offering guidance for their application in contemporary empirical research.Esta revisión examina los avances metodológicos recientes en la estimación de los efectos causales utilizando marcos de diferencias en diferencias (DiD), en particular bajo la adopción de tratamientos escalonados con efectos heterogéneos. Aunque los modelos de efectos fijos bidireccionales (TWFE) siguen siendo ampliamente utilizados, pueden producir resultados engañosos cuando los efectos del tratamiento varían entre unidades o a lo largo del tiempo. Para abordar estas preocupaciones, la revisión examina un conjunto de estimadores alternativos desarrollados para mejorar la identificación y la interpretación en tales contextos. Entre ellos se incluyen los estimadores dinámicos en tiempo de grupo, los métodos basados en la imputación y las estrategias doblemente robustas. El artículo complementa esta visión teórica con una serie de estudios de simulación diseñados para evaluar el rendimiento empírico de estos métodos en distintos grados de heterogeneidad del efecto del tratamiento y de complejidad de la estructura del error. Los resultados ponen de relieve tanto el potencial como las limitaciones de los modernos estimadores DiD, ofreciendo orientación para su aplicación en la investigación empírica contemporánea
El pirata de ficció: representació i evolució de la figura literària del pirata a "El pirata" (1822) de Walter Scott, "L’illa del tresor" (1883) de Robert Louis Stevenson, i "El pirata" (1923) de Joseph Conrad
No full textTreball de Fi de Grau en Humanitats. Curs 2024-2025Tutora: Maria Teresa Vinardell PuigLa figura del pirata ens desconcerta o ens fascina? Com es configura en la literatura i quins aspectes caracteritzen l’arquetip que hem creat en el nostre imaginari col·lectiu? Aquest Treball de Fi de Grau d’Humanitats titulat «El pirata de ficció. Representació i evolució de la figura literària del pirata a El pirata (1822) de Walter Scott, L’illa del tresor (1883) de Robert Louis Stevenson i El pirata (1923) de Joseph Conrad» analitza la figura literària del pirata i estudia com es construeix a través dels personatges Clement Cleveland, John Silver el Llarg i Jean Peyrol. Així, sense pretendre donar una resposta definitiva, l’objectiu d’aquest treball recau en analitzar els tres pirates literaris posteriors a l’Època daurada de la pirateria a partir de tres eixos que els interrelacionen: la marginalitat, l’amor i l’ambivalència, per tal de dur a terme una aproximació sobre la figura del pirata de ficció en tres obres de la literatura dels segles XIX i XX. D’aquesta manera, en un viatge al passat, el següent treball contribueix en la visió diacrònica i crítica d’una figura clau en la literatura d’aventures
Multilingualism in video games and L3 solutions across English, Spanish, Japanese, and Chinese localization
No full textAquesta tesi investiga el multilingüisme i les solucions de traducció per a L3 en videojocs mitjançant un marc interdisciplinari que integra estudis de jocs, teoria de la traducció, localització audiovisual i estudis de recepció. Combinant l’anàlisi del discurs de textos que són elements del joc amb estudis qualitatius de recepció en fòrums de jugadors, informes industrials i paratextos, la recerca examina tres videojocs —Cyberpunk 2077 (2020), Octopath Traveler (2018) i Baldur’s Gate 3 (2023)— per categoritzar les representacions de L3 (llengües naturals, variacions lingüístiques, i llengües inventades) i avaluar-ne les solucions de traducció en les versions anglesa, espanyola, japonesa i xinesa, molt especialment per a la combinació anglès-xinès. Amb l'anàlisi de les prioritats i les restriccions a cada adaptació de L3, l’estudi destaca disparitats en les relacions entre L1, L2 i L3 per a cada cas. A més, ressenya les dinàmiques a tres bandes entre desenvolupadors, traductors i jugadors, i proposa un model col·laboratiu per abordar els reptes sistèmics de localització. Més enllà de l’anàlisi lingüística, aquesta tesi proposa una metodologia interdisciplinària per a la recerca en l'àmbit de localització de videojocs, contribuint a debats acadèmics en traductologia i mitjans digitals al posar de manifest de quina manera el multilingüisme influeix en la immersió del jugador així com l'autenticitat cultural de les narratives interactives distribuïdes globalment.Esta tesis investiga el multilingü ismo y las soluciones de traducción para L3 en videojuegos mediante un marco interdisciplinario que integra estudios de juegos, teoría de la traducción/localización audiovisual y estudios de recepción. Combinando el análisis del discurso de textos intrajuego con estudios cualitativos de recepción en foros de jugadores, informes industriales y paratextos, la investigación examina tres videojuegos—Cyberpunk 2077 (2020), Octopath Traveler (2018) y Baldur’s Gate 3 (2023)—para categorizar las representaciones de L3 (lenguas naturales, variación lingüística y lenguas inventadas) y evaluar sus soluciones de traducción de versiones en inglés, español, japonés y chino, con atención especial a la combinación inglés-chino. Al analizar las prioridades y las restricciones de las adaptaciones de L3, el estudio destaca las disparidades en las relaciones ente L1, L2, y L3, para cada caso. Además, reseña las dinámicas tripartitas entre desarrolladores, traductores y jugadores, proponiendo un modelo colaborativo para abordar los desafíos sistémicos de localización. Más allá del análisis lingüístico este trabajo propone una metodología interdisciplinaria para la investigación de la localización de videojuegos, contribuyendo a debates académicos en estudios de traducción y medios digitales al dilucidar cómo el multilingüismo influye en la inmersión del jugador y la autenticidad cultural en narrativas interactivas distribuidas globalmente.This thesis investigates multilingualism and solutions for L3 as a translation problem in video games through an interdisciplinary framework integrating game studies, audiovisual translation/localization theory, and reception studies. Combining discourse analysis of in-game texts with qualitative reception studies of player forums, industry reports, and paratexts, the research examines three video games—Cyberpunk 2077 (2020), Octopath Traveler (2018), and Baldur’s Gate 3 (2023)—to categorize L3 representations (natural languages, language variations, constructed languages) and evaluate their translation solutions across English, Spanish, Japanese, and Chinese versions, with emphasis on English-Chinese contrasts. By analyzing priorities and restrictions in L3 adaptation, the study highlights disparities in L1/L2/L3 relationships. It critiques the tripartite dynamics among developers, translators, and players, proposing a collaborative model to address systemic localization challenges. Beyond linguistic analysis, this work advances an interdisciplinary methodology for video game localization research, contributing to scholarly debates in translation studies and digital media by elucidating how multilingualism shapes player immersion and cultural authenticity in globally distributed interactive narratives.Universitat Pompeu Fabra. Doctorat en Traducció i Ciències del Llenguatg
I tu, què mires?: les imatges i la cultura visual com a eina per a pensar el món des d'una perspectiva transfeminista i decolonial
No full textFanzine pedagògic que recull el resultat del projecte d’innovació docent per l'assignatura: Fonaments i Teoria de la Història de l'Art. Curs 2024-2025Projecte del Pla d’Ajuts de Suport a la Qualitat i a la Innovació en Aprenentatge i Coneixement de la Universitat Pompeu Fabra. Convocatòria PlaCLIK 2023-2024Aquest projecte d’innovació docent té com a objectiu incorporar tres aproximacions teòrico-pràctiques encara poc presents en el currículum acadèmic de les humanitats: la teoria crítica feminista, la perspectiva decolonial i la proposta metodològica d’Aby Warburg. Totes tres comparteixen una voluntat transformadora en la manera d’aproximar-se a l’art, trencant amb narratives tradicionals centrades en el geni creador, les biografies individuals, la mirada eurocèntrica i les categories universals de la modernitat il·lustrada.
A través de la metodologia de les Estratègies de Pensament Visual (VTS), l’alumnat desenvoluparà una capacitat crítica i creativa per interpretar imatges i objectes artístics, tot incorporant-hi reflexions sobre el gènere, la identitat, el cos i els afectes. L’art es concep com a eina de coneixement i anàlisi cultural, amb potencial per generar preguntes i propostes que desafien els relats dominants
A head-to-head comparison of plasma biomarkers to detect Alzheimer's disease in a memory clinic
No full textIntroduction: Blood-based biomarkers for Alzheimer's disease (AD) have been widely studied, but direct comparisons of several biomarkers in clinical settings remain limited. Methods: In this cross-sectional study, plasma biomarkers from 197 participants in the BIODEGMAR cohort (Hospital del Mar, Barcelona) were analyzed. Participants were classified based on AD cerebrospinal fluid (CSF) core biomarkers. We assessed the ability of plasma p-tau181, p-tau217, p-tau231, t-tau, and Aβ42/40 to classify Aβ pathology status. Results: Plasma p-tau biomarkers had a greater diagnostic performance and larger effect sizes compared to t-tau and Aβ42/40 assays in detecting Aβ pathology. Among them, plasma p-tau217 consistently outperformed the others, demonstrating superior area under the curves. Furthermore, p-tau217 showed the strongest correlation between plasma and CSF levels, underscoring its potential as a reliable surrogate for CSF biomarkers. Discussion: Several plasma biomarkers, targeting different epitopes and using different platforms, demonstrated high performance in detecting AD in a memory clinic setting. Highlights: Plasma p-tau biomarkers demonstrated higher diagnostic performance and larger effect sizes than t-tau and Aβ42/40 assays in detecting Alzheimer's disease. Among the p-tau biomarkers, p-tau217 assays consistently outperformed the others, providing superior classification of Aβ pathology status across different phosphorylation sites. p-tau217 assays showed the strongest correlation between plasma and CSF levels, indicating its potential as a reliable surrogate for CSF biomarkers. Several plasma p-tau biomarkers can be used in a specialized memory clinic to accurately detect Alzheimer's disease.Federica Anastasi receives funding from the JDC2022-049347-I grant, funded by the MCIU/AEI/10.13039/501100011033 and the European Union NextGenerationEU/PRTR. Henrik Zetterberg is a Wallenberg Scholar and a Distinguished Professor at the Swedish Research Council supported by grants from the Swedish Research Council (#2023-00356; #2022-01018 and #2019-02397), the European Union's Horizon Europe research and innovation programme under grant agreement No 101053962, Swedish State Support for Clinical Research (#ALFGBG-71320), the Alzheimer Drug Discovery Foundation (ADDF), USA (#201809-2016862), the AD Strategic Fund and the Alzheimer's Association (#ADSF-21-831376-C, #ADSF-21-831381-C, #ADSF-21-831377-C, and #ADSF-24-1284328-C), the European Partnership on Metrology, co-financed from the European Union's Horizon Europe Research and Innovation Programme and by the Participating States (NEuroBioStand, #22HLT07), the Bluefield Project, Cure Alzheimer's Fund, the Olav Thon Foundation, the Erling-Persson Family Foundation, Familjen Rönströms Stiftelse, Stiftelsen för Gamla Tjänarinnor, Hjärnfonden, Sweden (#FO2022-0270), the European Union's Horizon 2020 research and innovation programme under the Marie Skłodowska-Curie grant agreement No 860197 (MIRIADE), the European Union Joint Programme – Neurodegenerative Disease Research (JPND2021-00694), the National Institute for Health and Care Research University College London Hospitals Biomedical Research Centre, and the UK Dementia Research Institute at UCL (UKDRI-1003). Marc Suárez-Calvet receives funding from the European Research Council (ERC) under the European Union's Horizon 2020 research and innovation programme (Grant agreement No. 948677); ERA PerMed-ERA NET and the Generalitat de Catalunya (Departament de Salut) through the project SLD077/21/000001; Project “PI19/00155″ and “PI22/00456, funded by Instituto de Salud Carlos III (ISCIII) and co-funded by the European Union; and from a fellowship from ”la Caixa” Foundation (ID 100010434) and from the European Union's Horizon 2020 research and innovation programme under the Marie Skłodowska-Curie grant agreement No 847648 (LCF/BQ/PR21/11840004)