Optimising Computational Methods in B Cell Lymphoma

Abstract

Diffuse large B-cell lymphoma (DLBCL) is a remarkably heterogeneous disease, presenting significant challenges in the classification, monitoring, and treatment of patients. Throughout my PhD, I set out to address some of these challenges by developing and applying novel computational methods to enhance the robustness and interpretability of high-throughput sequencing data in lymphoma research. This thesis is structured around three interlinked projects. First, I introduce noisyR, an R package designed to systematically quantify and remove technical noise from sequencing datasets. By focusing on consistency of signal across samples, noisyR helps ensure that downstream analyses like differential expression, enrichment, and network inference are more consistent and interpretable. The package is complemented by bulkAnalyseR, an interactive pipeline that lowers the barrier for researchers to perform robust transcriptomic analysis, supporting open and reproducible science. Second, I present the results of the DIRECT clinical study of aggressive B-cell lymphoma, which assessed the feasibility of using circulating tumour DNA to monitor patient disease. By implementing advanced error suppression strategies, I helped develop a robust pipeline to perform patient risk stratification, molecular subtyping, and minimal residual disease monitoring in a real-world clinical setting. Third, I describe a large-scale single-cell perturbational transcriptomics experiment which systematically introduced over 300 wild-type and mutant open reading frames into healthy primary human germinal centre B cells. By combining innovative barcode assignment and novel correlation-based methods for analysis, I characterised the strength and transcriptional consequences of recurrent lymphoma mutations, both reproducing existing knowledge and uncovering novel functional relationships. Taken together, the results of this thesis advance computational methodology for sequencing data analysis and demonstrate how these tools can generate new biological insights into DLBCL. The approaches developed here improve data quality, reproducibility, and interpretability, and I hope they will lay the groundwork for future translational applications in precision oncology and personalised medicine.