53417 research outputs found
Sort by
Linguistic ethnography for digital inclusion: Adapting digital platforms with and for autistic people
No full textP104: BASE EDITING REPAIRS THE HBE MUTATION RESTORING THE PRODUCTION OF NORMAL GLOBIN CHAINS IN SEVERE HBE/β‐THALASSEMIA PATIENT HEMATOPOIETIC STEM AND ERYTHROID CELLS
No full textHarmonising definitions of multiple long term conditions for inflammation research: a co-production approach.
Get PDFStructured Approaches for Optical Music Recognition Integrating Enhanced Datasets, Specialised Architectures and Multi-level Evaluation
Get PDFOptical Music Recognition (OMR) aims to convert images of musical scores into machine-readable representations, enabling digital preservation, accessibility and computational analysis of musical works. OMR remains challenging due to the complex, context-dependent nature of musical notation, which encodes information through diverse symbols with varying visual characteristics and intricate spatial relationships. This thesis presents an approach to advancing OMR through four related contributions. First, it introduces DoReMi and DoReMi v2, novel datasets comprising over 15,000 high-resolution score images with nearly 3.3 million annotated objects across 217 symbol classes. These datasets uniquely provide both agnostic and semantic representations, explicit rhythmic positions and structural annotations that support all stages of the OMR pipeline. Second, this research investigates specialised architectural approaches for different aspects of musical notation recognition. Through experiments, it establishes that two-stage detectors excel at geometric primitives like staff lines and barlines, while single-stage detectors better recognise complex symbols like noteheads and dynamics. Instance segmentation provides pixel-level precision needed for overlapping symbols and domain-specific preprocessing enhances recognition robustness. Third, the thesis develops a knowledge-driven late fusion architecture that combines these strengths. This approach integrates symbol-specific performance differentials into a principled fusion algorithm, achieving significantly improved recognition accuracy compared to individual model approaches. Fourth, the research formalises a multi-dimensional evaluation framework that builds on and integrates existing notions of symbol-, structure- and semantics-aware metrics into a hierarchical methodology. While prior work has addressed these dimensions in isolation, this framework aligns them with musical abstraction levels, symbolic, structural, semantic and usability, allowing analysis of how low-level recognition affects higher-level musical understanding and practical utility. By operationalising these layers through dedicated metrics such as Edge F1 and Musical Edit Distance, the framework presents a replicable and musically informed standard for OMR system evaluation. Experiments show that different detection architectures exhibit consistent performance differences across musical symbol categories, with performance differentials reaching 46.06%. The integration of specialised detection components via informed fusion yields performance improvements that show the benefit of using specialised architectures and suggest this approach could extend to other recognition problems
Association of comorbidities and socioeconomic deprivation among people who died from dementia in England Between 2013-2023: analysis of death certificates
No full textSynaptic and intrinsic membrane defects disrupt early neural network dynamics in Down syndrome.
Get PDFDown syndrome, caused by trisomy 21, affects around six million people worldwide and features learning, memory and language deficits. However, the mechanisms underlying trisomy 21 neurophenotypes involving human cortical circuitry are unknown. By characterising developing neural network dynamics and single cell excitability profiles, from synaptic and voltage-dependent ion channel behaviour using an isogenic induced pluripotent stem cell-derived neuronal model, we show that trisomy 21 impairs the activity and development of cortical circuitry. This is caused by deficient glutamatergic synaptic connectivity and by aberrant intrinsic membrane properties involving K+ and Na+ channels culminating in spike firing defects that weaken neural network activity and disrupt the synchrony of developing neurons. We also identify transiently activated A-type K+ channels, specifically Kv4.3 channels, as a key orchestrator for Down syndrome during neurodevelopment. Overall, these excitability changes will significantly contribute towards the aberrant neurophenotypes observed later on in life
Advancements in Liquid-phase Peptide Synthesis by Nanostar Sieving
No full textPeptide drugs make up a significant and increasing portion of the pharmaceutical market. As a result, peptide synthesis is a thriving field of academic study. There is a strong motivation within the pharmaceutical industry to reduce the cost of peptide synthesis to improve profit margins and make peptide drugs more affordable, broadening the potential customer base. Solid-phase synthesis has historically been the most common method for the chemical synthesis of peptides; however, due to the large amounts of reagents and solvents required it can be costly to scale, hence there are many alternative methods currently being explored. Previously the liquid-phase peptide synthesis by nanostar sieving (PEPSTAR) method has been reported in which peptides up to 10 amino acids in length were synthesized on a multi-armed uniform-mass polyethylene glycol (PEG) support referred to as a nanostar, with the reagents and byproducts of each coupling cycle removed by organic solvent nanofiltration (OSN). OSN is particularly attractive to the pharmaceutical industry as a purification process as it is scalable, simple to operate, and easy to automate. This thesis describes advancements in the PEPSTAR method including: i) the simplification of the PEPSTAR synthesizer from two-stage to single-stage diafiltration enabled by the selection of a new membrane/nanostar combination, with nanostar rejections ranging from 99.4 – 100 % ; ii) the first isolation of peptides from the nanostar debris by OSN, resulting in 68 % yield of a decamer across the entire synthesis; iii) the use of a dual photolabile/acid-labile linker strategy for the convenient monitoring of PEPSTAR; iv) the synthesis of a 39-mer peptide by the assembly of peptide fragments in 88 % yield, with fragment excess removed by OSN after each coupling cycle; v) the use of NMR to quantify peptide intermediates in the PEPSTAR synthesizer; vi) the use of green solvents in PEPSTAR, and vii) an assessment of the process mass intensity (PMI) of PESPTAR and a reduction in PMI by increasing the concentration at which synthesis is performed, with dilution of the synthesis only when gelation becomes an issue. The sum of this research constitutes significant advancements in the PEPSTAR methodology, and progress towards the use of PEPSTAR for the large-scale synthesis of peptides in an industrial setting
