Publikationer från Uppsala Universitet
Not a member yet
127603 research outputs found
Sort by
University students learning astronomy through observations at professional telescopes : A social semiotic case study
No full textFormalising Anti-Discrimination Law in Automated Decision Systems
No full textAlgorithmic discrimination is a critical concern as machine learning models are used in high-stakes decision-making in legally protected contexts. Although substantial research on algorithmic bias and discrimination has led to the development of fairness metrics, several critical legal issues remain unaddressed in practice. The paper addresses three key shortcomings in prevailing ML fairness paradigms: (1) the narrow reliance on prediction or outcome disparity as evidence for discrimination, (2) the lack of nuanced evaluation of estimation error and assumptions that the true causal structure and data-generating process are known, and (3) the overwhelming dominance of US-based analyses which has inadvertently fostered some misconceptions regarding lawful modelling practices in other jurisdictions. To address these gaps, we introduce a novel decisiontheoretic framework grounded in anti-discrimination law of the United Kingdom, which has global influence and aligns closely with European and Commonwealth legal systems. We propose the "conditional estimation parity" metric, which accounts for estimation error and the underlying data-generating process, aligning with UK legal standards. We apply our formalism to a real-world algorithmic discrimination case, demonstrating how technical and legal reasoning can be aligned to detect and mitigate unlawful discrimination. Our contributions offer actionable, legally-grounded guidance for ML practitioners, policymakers, and legal scholars seeking to develop non-discriminatory automated decision systems that are legally robust
Spatial TCR clonality and clonal expansion in the in situ microenvironment of non-small cell lung cancer
No full textBackground T-cell activation and clonal expansion are essential to effective immunotherapy responses in non-small cell lung cancer (NSCLC). The distribution of T-cell clones may offer insights into immunogenic mechanisms and imply potential prognostic and predictive information. Methods We analyzed alpha/beta T-cell receptor (TCR) clonality using RNA-sequencing of bulk frozen tumor tissue from 182 patients with NSCLC. The data was integrated with molecular and clinical characteristics, extensive in situ imaging, and spatial sequencing of the tumor immune microenvironment. TCR clonality was also determined in an independent cohort of nine patients with immune checkpoint-treated NSCLC. Results TCR clonality (Gini index) patterns ranged from high T-cell clone diversity with high evenness (low Gini index) to clonal dominance with low evenness (high Gini index). Generally, TCR clonality in cancer was lower than in matched normal lung parenchyma distant from the tumor (p=0.021). The TCR clonality distribution between adenocarcinoma and squamous cell carcinoma was similar; however, smokers showed a higher Gini index. While in the operated patient with NSCLC cohort, TCR clonality was not prognostic, in an immune checkpoint inhibitor-treated cohort, high TCR clonality was associated with better therapy response (p=0.016) and prolonged survival (p=0.003, median survival 13.8 vs 2.9 months). On the genomic level, a higher Gini index correlated strongly with a lower frequency of epidermal growth factor receptor (EGFR) and adenomatous polypsis coli (APC) gene mutations, but a higher frequency of P53 mutations, and a higher tumor mutation burden. In-depth characterization of the tumor tissue revealed that high TCR clonality was associated with an activated, inflamed tumor phenotype (PRF1, GZMA, GZMB, INFG) with exhaustion signatures (LAG3, TIGIT, IDO1, PD-1, PD-L1). Correspondingly, PD-1+, CD3+, CD8A+, CD163+, and CD138+immune cells infiltrated cancer tissue with high TCR clonality. In situ sequencing recovered single dominant T-cell clones within the patient tumor tissue, which were predominantly of the CD8 subtype and localized closer to tumor cells. Conclusion Our robust analysis pipeline characterized diverse TCR repertoires linked to distinct genotypes and immunologic tumor phenotypes. The spatial clustering of expanded T-cell clones and their association with immunological activation underscores a functional, clinically relevant immune response, particularly in patients with NSCLC treated with checkpoint inhibitors
Classifying Hate : Legal and Ethical Evaluations of ML-Assisted Hate Crime Classification and Estimation in Sweden
No full textHate crimes, driven by biases against specific demographic groups, harm not only individuals but undermine the security, trust, and cohesion of entire communities. Accurately identifying such crimes remains a significant challenge due to under-reporting, limited training, and the complexity of determining bias motivations. In this paper, we analyze the results of a text classification model developed to improve the precision of hate crime statistics and identification in Sweden. Empirical results indicate the model outperforms traditional manual police classification of hate crimes, achieving higher precision across various crime types and regions. We further disaggregate performance to pinpoint persistent challenges and highlight categories where both human and machine decision-makers struggle. While the model focuses on statistical estimation rather than direct case-level decision-making, we discuss the broader implications of algorithmic transparency, accountability, and explainability. Ultimately, this research illustrates how transformer-based neural networks can responsibly bolster the detection and understanding of hate crimes, informing policies to better protect vulnerable communities
Deciphering the molecular landscape of childhood acute lymphoblastic leukemia through multiomics data integration
No full textPediatric acute lymphoblastic leukemia (ALL), the most common childhood cancer, is characterized by aberrant lymphopoiesis that disrupts normal hematopoiesis in the bone marrow. Diverse genetic drivers, and epigenetic and transcriptomic alterations contribute to disease heterogeneity. By applying standard clinical diagnostic procedures and next generation sequencing (NGS) methods, 22 B-cell precursor ALL (BPC-ALL) and 15 T-cell ALL (T-ALL) subtypes have been identified. The aim of this thesis is to integrate molecular, clinical, and ex vivo drug response data from pediatric ALL patients diagnosed and treated in the Nordic countries to identify cross-modal interactions underlying therapy responses and clinical outcomes. Ultimately, this work aims to provide insights that support treatment optimization, with the dual goal of preventing relapses and reducing toxicity. The datasets included DNA methylation (DNAm), gene expression (GEX), somatic single nucleotide variants (SNVs), fusion genes, copy number alterations (CNA), ex vivo drug response (EVDR) data, and clinical variables from over 1,000 patients. Bioinformatic approaches included supervised classification, survival analysis and modelling, differential DNAm and GEX analyses, and data integration. In paper I, we built ALLIUM, a multimodal machine learning classifier that predicts 17 ALL subtypes using either DNAm or GEX data. Applied to 1,131 ALL patients, ALLIUM successfully subtyped previously unclassified cases, usually aligning with existent molecular findings. In paper II, we designed two DNAm-based risk prediction models, the relapse risk predictor (RRP) and the mortality risk predictor (MRP), which outperformed the traditional risk stratification models and generalized well to independent cohorts. In paper III, we integrated EVDR data with DNAm and GEX data and found pre-treatment epigenetic and transcriptomic differences between ex vivo resistant and sensitive BCP-ALL patients. The resulting signatures were enriched in pathways commonly altered in cancer. In paper IV, we utilized a framework to integrate DNAm, GEX, SNV, and EVDR data from 1,231 BCP-ALL patients. We identified cross-modal signatures and associations, reducing feature dimensionality and enabling pathway, correlation, and network analyses. These signatures improved risk modelling beyond standard clinical stratification. Together, these studies enhance our understanding of the molecular underpinnings of ALL and may form a basis for supporting future clinical decision-making, particularly in ambiguous cases
Projected external doses from an accidental release of ESS spallation-target products : time-dependence and radionuclide contribution
No full textThe Swedish Radiation Safety Authority has presented a report on severe accident scenarios at the European Spallation Source (ESS) for dimensioning the emergency preparedness zones around the facility. The source-term in the scenario consisted of more than 80 tungsten-target spallation products with physical half-life (T½) exceeding 1 hour. The purpose of this study is to establish which of these radionuclides will become of highest importance in terms of the radiological consequences to residents in areas affected by an accident release. In analogy with accidents at nuclear power plants, where the fission product 137Cs is a key nuclide for estimating projected external doses to affected residents, a corresponding key nuclide for ESS is required for the emergency preparedness. Using existing accident source terms in combination with reported values on ecological half-times of the gamma emitter 137Cs, the external dose rates and cumulative doses per unit initial ground deposition of the suggested key-nuclide (182Ta) could be estimated. In terms of 50 y dose from a dry deposition of the released source-term, 172Lu (T½ = 6.7 d, supported by 172Hf with T½ = 1.87 y) contributes up to 50% of the 50 y dose, depending on the ecological half-times for the element. The isomer 178nHf (T½ = 31 y) is the second largest contributor to gamma-ray dose, followed by 182Ta (T½ = 115 d), that contribute with about 15% and 10% of the 50 y dose, respectively. The results thus suggest that 172Hf/172Lu may be more suitable for long-term follow-up of projected doses from accidental ESS releases than 182Ta
Differential effects of freshwater browning across fish species : consequences for individual- to community-level fish traits in north temperate lakes
No full textThe browning of freshwater ecosystems is increasingly evident in temperate and northern regions, with widespread ramifications for lake physics, chemistry, and biology. Contrasting results on how freshwater browning may impact fish have been reported, but there has been no comprehensive examination of how browning may cause cascading effects on individual- to population- to community-level traits of freshwater fishes. We addressed this knowledge gap by summarizing the existing literature and conducting a series of original analyses to: (i) explore the effects of a brown water gradient on populations of eight economically important species of fish across 871 lakes; and (ii) examine how a brown water gradient may influence community trait compositions across 303 lakes. From our literature synthesis, we found that fish growth is often negatively associated with browner waters, despite browning generally showing no effect on fish foraging. We also demonstrated that browner waters had greater abundances of northern pike (Esox lucius) and walleye (Sander vitreus), but lower numbers of lake trout (Salvelinus namaycush), yellow perch (Perca flavescens), largemouth bass (Micropterus salmoides), smallmouth bass (M. dolomieu), and lake whitefish (Coregonus clupeaformis). Moreover, we showed that fish communities were significantly more likely to contain species with larger eyes in browner lakes. Lastly, we examined relationships between various metrics of browning (i.e. dissolved organic carbon, Secchi transparency, water colour) and present a framework for how the effects of freshwater browning on fish may scale from individuals to populations to communities
Modelling and Simulation of Real-time Robotics Software in MIMOS
No full textThis thesis presents a series of real-time robotic applications utilising the MIMOS framework, a deterministic, timed programming model designed explicitly for safety-critical systems. It begins with a description of the MIMOS functional model. Several benchmark algorithms from the RTRBench suite are translated into MIMOS-based implementations by decomposing them into functional nodes with defined timing and communication constraints. To validate these implementations, simulation results obtained from the MIMOS framework are compared with those reported in the original studies. Representative case studies include extended Kalman filter-based SLAM, symbolic planning, and dynamic movement primitives, among others. Additionally, this thesis examines integration strategies for connecting MIMOS with external simulation tools and programming environments, such as Modelica and Simulink. Several approaches are proposed to facilitate the incorporation of widely used simulators into the MIMOS workflow, thereby extending its applicability to hybrid systems and complex scenarios
Ups and downs of the Guiana Shield and Amazon Basin over the last 500 Myr
No full textCratons such as the Guiana Shield are often considered as stable regions, undergoing long-term emergence and denudation due to buoyancy. However, by integrating geological and geomorphological observations with apatite fission-track analysis, we define a history involving repeated episodes of burial and exhumation over the last 500 Myr. Over much of the shield, the thermal history is dominated by the effects of earliest Jurassic magmatism, followed by Early Cretaceous exhumation coincident with the onset of seafloor spreading in the southern South Atlantic. Further episodes of regional exhumation occurred in Aptian-Albian time coincident with a global-scale plate reorganisation and in Eocene times coincident with a slowdown in the movement of the South American plate. Results from the Amazon Basin also define these four episodes. Thermal data from a deep well in the Amazon Basin show that the Early Cretaceous and Eocene exhumation episodes were preceded by burial by kilometre-scale thicknesses of cover, subsequently removed. Continuity of data from basin to shield suggests that burial extended across the shield. Early Cretaceous exhumation led to formation of a baseCretaceous peneplain across the entire continent, from the Andes (during post-orogenic collapse) to the Amazon Basin and the Guiana Shield. This peneplain was then buried beneath Cretaceous-Paleogene sediments (remnants of which are preserved in the Amazon well) prior to the onset of Eocene exhumation, which also extended into in the offshore Casipore<acute accent> Basin (based on our results from a deep well). The Eocene episode also correlates with post-orogenic collapse of the Andes. The history of repeated burial and exhumation defined for the Guiana Shield appears to be a common property of supposedly stable cratons. The correlation between Andean tectonics, episodes of exhumation defined here and changes in the motion of the South American plate, shows that plate-tectonic changes governed the vertical movements across the continent
Geotermisk potential i Sundhnúkurs nybildade sprickgångar på Reykjaneshalvön
No full textGeothermal energy is an attractive way of energy production where it can be achieved. Iceland ranks among the world’s leaders in its utilization. Svartsengi, a geothermal powerplant on the Reykjanes Peninsula in SW-Iceland, has produced electricity and hot water for a few decades. The recent Sundhnúkur dike intrusions/eruptions close to Svartsengi have now created new opportunities for geothermal development. This thesis assesses the scale and persistence of the newly added heat and its implications for siting and timing future development near Svartsengi. Geothermal prospects associated with the recent Sundhnúkur dike sequence on the Reykjanes Peninsula are evaluated to learn the magnitude, footprint, and lifetime of intrusion-driven heat. A 2-D Finite-Element heat transfer model integrates conductive heat transfer with Darcy advection in a permeable basaltic crust, following the observed intrusion timeline. Boundary conditions fixed 5 °C at the surface and 500 °C at 5 km depth; dikes are initialized at 1240 °C, and the system commenced from 30 years prior to 50 years after the final intrusion. Fifteen years post emplacement, thermal anomalies remain discrete and centered around the dikes. By 30 years they broaden and coalesce while peak amplitudes decline. Depth profiles show excess temperatures of 15–20 °C at 500–750 m, increasing to 40–60 °C at 1000m and 75–85 °C at 2000– 3000 m depth. Warming is strongest within less than 40 m laterally of dikes and persists longest across narrow gaps, indicating proximity and spacing as primary controls. Based on the findings of this thesis, wells drilled into this area for geothermal exploration should be drilled after 15-20 years and sited as close as safely practicable (0-40 m) to the widest dikes at the shallowest depth meeting this threshold. Main limitations of the model are the 2-D geometry, homogenous crust and omission of the two most recent intrusions. Orkuframleiðsla með jarðhita er mikill kostur þar sem aðstæður leyfa. Ísland er meðal fremstu þjóða í þróun og nýtingu jarðvarma og jarðvarmavirkjunin á Svartsengi hefur framleitt rafmagn og heitt vatn í nokkra áratugi. Nýleg innskot og eldgos við Sundhnnúk skammt frá Svartsengi hafa skapað ný tækifæri til frekari jarðhitanýtingu. Í þessari ritgerð er umfang og stöðugleiki hins nýja varma metinn ásamt tímasetningu og mögulegrar staðsetningar til frekari framkvæmda. Framtíðarmöguleikar Sundhnúkagígaraðarinnar til orkunýtingar eru greindir með því að ákvarða stærð, útbreiðslu og líftíma varmans frá innskotunum. Sett var upp tvívíddarlíkan byggt á reiknilíkani með bútaaðferð með lögmáli Darcy í gropnu basísku bergi miðað við rauntíma eldvirkninnar. Í líkaninu var yfirborðshiti 5 °C og 500 °C á 5 km dýpi en innskotins við 1240 °C. Líkanið nær yfir tímabil sem spannar frá 30 árum fyrir fyrsta innskot til 50 árum eftir lokainnskot. Niðurstöður sýna að um 15 árum eftir myndun er varminn frá hverju innskoti enn vel afmarkaður en eftir 30 ár hefur hann breiðst út og birtist frekar sem ein eining en átta aðskildum innskotum. Dýptarprófílar sýna fram á um 15-20°C hlýnun í 500-750 m dýpi, 40-60°C á 1000 m dýpi og 75-85°C á 2000-3000 m dýpi. Hlýnun er mest innan við 40 m frá innskotunum og varir lengst þar sem þröngt er á milli innskotanna. Á grundvelli þessara niðurstaðna er mælt með því að rannsóknarholur verði boraðar eins nálægt innskotunum og unnt er og á grynnsta dýpi sem uppfyllir tilgreint hitamarkmið eftir um 15-20 ár.