Epstein-Barr Virus (EBV), a ubiquitous herpesvirus that infects over 95% of the adult human population, has been implicated in the aetiology of a range of autoimmune diseases and tumours. In some of these disorders such as post-transplant B-cell lymphomas, EBV acts as a direct causal factor, in others, like Hodgkin's disease and nasopharyngeal carcinoma, it is an important co-factor. Additionally, EBV infection has been linked to several other diseases, most notably Multiple Sclerosis through positive correlation with the occurrence of Infectious Mononucleosis – a benign lymphoproliferative disease caused by primary EBV infection. The key feature of most EBV-disease associations is the ability of the virus to infect and transform human B- T- NK- and epithelial cells using a set of transcripts and proteins, some of which act as oncogenes. While it is evident that EBV viral load and gene expression may be correlated with the course of disease or even directly contributing to its pathology, the genetic determinants of EBV uptake, expression and its proliferative capacity remain unresolved. This project aimed to investigate the genetic determinants of EBV copy number and EBV latency gene expression for human B-cells immortalised by EBV in vitro and transformed into permanently growing lymphoblastoid cell lines (LCLs), as a model for early-stage EBV infection in naïve B-cells. LCL samples studied have been sourced from several different populations, the HapMap Project, the 1000 Genomes Project as well as British MRC-A family cohort. Methods used encompass quantification of viral expression and copy number using TaqMan and SybrGreen PCR techniques, followed by statistical association tests conducted using Plink, Merlin and MatrixEQTL. EBV QTLs identified by the assays were next subjected to a meta-analysis in GWAMA. Two most significant eQTLs were also selected for a replication experiment in an independent panel of newly generated LCLs and validated in peripheral blood B-cells sourced from the same donors. Multiple significant and suggestive expression and copy number QTLs were identified. However, most of these associations have not been replicated in more than a single cohort. The relatively small sample size of most cohorts tested as well as population structure posed a limitation. Some findings merit attention, particularly the presence of statistically significant viral eQTLs within or close to CSMD1 locus in two different cohorts, and finding of a significant EBV eQTL in a SNP associated with type 1 diabetes risk and located close to IL2RA, an immune-response gene harbouring multiple autoimmune disease risk loci. Suggestive associations were also identified in the 1000 Genomes Project samples by the copy number assay which resulted in the most robust test conducted. These encompassed an association to the PRDM9 locus as well as to a gene involved in TGF-β secretion. This is particularly interesting since TGF-β signal promotes lytic replication in EBV-infected B-cells and a consistent significant correlation between EBV lytic expression and increased viral copy number has been identified. In conclusion, although no significant association has been consistently replicated, the project provided several suggestive EBV QTL candidates with plausible biological links to EBV infection and replication, which could be studied further in independent experiments.

No description available.

Recent improvements in deep learning architectures, combined with the strength of modern computing hardware such as graphics processing units, has lead to significant results in the field of image analysis. In this thesis work, locally connected architectures are employed to reduce noise in flash X-ray diffraction images. The layers in these architectures use convolutional kernels, but without shared weights. This combines the benefits of lower model memory footprint in convolutional networks with the higher model capacity of fully connected networks. Since the camera used to capture the diffraction images has pixelwise unique characteristics, and thus lacks equivariance, this compromise can be beneficial. The background images of this thesis work were generated with an active laser but without injected samples. Artificial diffraction patterns were then added to these background images allowing for training U-Net architectures to separate them. Architecture A achieved a performance of 0.187 on the test set, roughly translating to 35 fewer photon errors than a model similar to state of the art. After smoothing the photon errors this performance increased to 0.285, since the U-Net architectures managed to remove flares where state of the art could not. This could be taken as a proof of concept that locally connected networks are able to separate diffraction from background in flash X-Ray imaging.