Folate-mediated one-carbon metabolism is thought to play an important role in colorectal cancer. However, results from conventional analyses have been equivocal. We can improve upon these analyses by incorporating into the statistical analysis the underlying biologic knowledge. We propose to use a hierarchical modeling framework to investigate genetic variants and compare these findings to a stochastic search variable selection algorithm. The hierarchical modeling approach uses a first-stage conditional logistic regression to independently estimate the association between each polymorphism and risk. In contrast, the model selection approach allows for joint modeling by searching over the entire model space to identify single polymorphisms and combinations that may be influential. Common to both approaches is a secondstage model regressing the first-stage estimates on prior covariates specifying gene-specific characteristics. To demonstrate these approaches, we analyze 524 SNPs genotyped on 28 genes involved in the folate and DNA repair pathways in the Colon Cancer Family Registry. In silico priors are obtained using a mathematical model of one-carbon metabolism that uses folate enzyme kinetics and regulatory mechanisms to simulate the impact of genetic and nutritional variation on the entire pathway. We present results from the applied analysis and discuss potential advantages and limitations to each approach.
2
The use of genome-wide eQTL associations to identify novel genetic pathways involved in complex traits