Computational analysis of patterns of DNA damage

Cancer and bacterial infection are major killers across the world. Personalised cancer therapies are needed that respond to individual mixes of DNA damage, and new antibiotics are needed to r espond to resistance brought about by genetic mutations in bacteria [1] [2]. In this thesis, I analyse the gene sequence patterns next to small mutations in cancer cells, identifying those associated with substitutions and indels. I show that in the exome there is an excess of in-frame indels compared to frameshift mutations; evidence of negative selection. Next I analyse the associations between the sequences of driver genes and mutational frequency in cancers. I find most driver genes are more frequently mutated in those cancers where the re is a good match between the mean mutational fingerprint for that cancer and the fingerprint formed from the mutations found in the driver gene in question. I then extend existing work on mutational signatures, to identify novel bacterial mutational signatures. By comparing the signatures with those of human cancers and environmental mutagens, I identify alkylation as a driver of bacterial mutagenesis. Next I review translational drug discovery, highlighting the use of bioinformatics to identify drug targets and biomarkers, assess protein druggability; and predict opportunities for drug repositioning. Finally, I identify therapeutically actionable mutually exclusive gene pairs within human cancers. I show that the Poisson binomial distribution is better for identifying mutual exclusivity. The predictions are available on the new MexDrugs website, and my python implementation of the Poisson binomial test can be installed via pip.