The aim of this proposal is to specify formulae for allelic scores that index levels of expression, methylation and the metabolome and subsequently construct these scores in a large sample of AS cases and controls with GWAS data, and test to see whether the scores correlate with affection status, potentially identifying interesting biological pathways.

This analysis involves two distinct stages:(i) Genome-wide association analysis of genome-wide methylation, expression and metabolomic data. This will involve association analysis of both common and low frequency variants derived from genome-wide SNP chip platforms (Illumina 550K or 660K) and low density genome-wide sequencing, as well as imputed variants from the thousand genomes and UK10K projects. Because of computational constraints associated with the extremely large number of phenotypes modelled, statistical analysis will proceed using SNPTEST on a cleaned set of 8365 unrelated individuals of confirmed British ancestry. (ii) Construction of allelic scores to serve as genetic proxies for these molecular variables. I will investigate how allelic scores can best be constructed using a variety of approaches from simple weighted counts of strongly associated variants, to genome-wide allelic scores- an approach which I have pioneered in the context of individual risk prediction, to more sophisticated methods involving machine learning and lasso regression. The predictive validity of these measures will be assessed by cross validation, and/or using them to predict the same outcome in cohorts with similar data (e.g. TwinsUK in the case of genome-wide expression; Kettunen et al. (2012) in the case of metabolomic data). Please note, that this part of the project is pretty much identical to that proposed in MRC Unite Programme 4. Once the formulae for generating allelic scores has been validated, these scores will be derived and correlated with a large sample AS cases and controls with GWAS data.