Using big data and statistics to understand melanoma skin cancer

Human pigmentation, defined as the coloration of external tissues on the body, can be measured in many different ways and is highly heritable. The extent to which the genetic factors underlying one pigmentary measure influence another pigmentary measure is largely unknown. Pigmentation is one of the strongest risk factors melanoma and many of the genetic loci associated with risk of melanoma are known to be related to pigmentation. Thus it is hoped that insights into the genetics of pigmentation may improve our understanding of melanoma. It is reported in this thesis the results of genome-wide association analyses of the 350,000 UK Biobank participants for five pigmentary measures and three combined pigmentation scores. 500 signals from 322 loci associated are identified; 109 of these loci are novel. By generating polygenic risk scores (PRS) for four pigmentary measure and a combined score, it was demonstrated the utility these PRS have when modelling melanoma risk as they performed similar to melanoma-based PRS. Constructing joint-analysis pipelines that supplement current melanoma GWAS summary statistics with large-scale pigmentation-based GWAS summary statistics uncovered 88 potential novel melanoma loci. The discovery of many new loci improves our understanding of how the genetic architecture of melanoma relates to pigmentation. The results in this thesis increase our understanding of the genetic architectures of distinct pigmentary measurements by identifying 109 novel loci; how the genetics of pigmentary characteristics overlap; provide evidence of the equivalency of using pigmentation-derived PRS (AUC=0.64) to a melanoma-specific PRS (AUC=0.65) when modelling melanoma risk; and the viability of jointly-analysing melanoma with known risk pathway phenotypes to identify novel melanoma loci, highlighting that large gains in our understanding of the genetic architecture of melanoma may still be achieved with larger melanoma data sets and by supplementing with known risk pathway phenotypes