Big Data Approaches in Genetics

The field of human genetics has advanced rapidly with significant growth in data production and analysis. Modern genotyping chips can assess up to a million genetic variants per individual, and imputation methods can extend this to 10 million variants. Historically, genetic studies faced challenges due to limited sample sizes, which restricted analysis to single variant tests. However, the advent of large genetic consortiums and biobanks with extensive participant data has enabled comprehensive multivariant analyses. Genome-wide association studies (GWAS) aim to find links between genetic variants and traits by testing many genetic markers across the genome. These studies often involve complex statistical methods due to the high volume of data and traits analysed. GWAS has identified numerous significant genetic associations for various traits and diseases, contributing to understanding biological causality, heritability, and clinical risk prediction. In this thesis, I have developed genotype-based prediction models for human pigmentation using diverse and multi-ethnic training cohorts. The models were applied to ancient human samples, addressing the challenges of degraded and low-coverage ancient DNA. The approach involved training on densely genotyped samples and predicting on pseudo-haploid samples using weighted ridge regression models. Additionally, the study explores using multivariate statistical models to analyse survival in clinical data from melanoma patients. Using Cox proportional hazards models, the analysis identifies prognostic factors for survival and metastasis, highlighting the role of specific genetic markers in improving survival rates. These findings align with existing data from global sources. I have also built an R package for visualising GWAS results with Manhattan and Q-Q plots, optimised for handling large datasets efficiently. The tools and methods developed here demonstrate the potential of integrating big data analytics in genetic research and clinical applications. In this PhD dissertation, I have developed several multivariate statistical models like ridge regression and Cox Proportional Hazards in diverse sets of data. I have also designed novel algorithms and created statistical software packages that facilitate the handling of big data. In conclusion, I have worked on an eclectic combination of theoretical statistics, applied statistics, and computational methods in my objective of exploring big data approaches in genetics.