Proceedings of the 2013 annual meeting of the Netherlands Epidemiology Society
Volume 1 Issue S1 Abstract 52
M.G.M. Braem, Julius Center, UMC Utrecht, Utrecht, the Netherlands
D. Gilbert-Diamond, Dartmouth Medical School, Hanover, USA
T. Hu, Geisel School of Medicine, Dartmouth College, Lebanon, USA
P. Andrews, Geisel School of Medicine, Dartmouth College, Lebanon, USA
P.H.M. Peeters, Julius Center, UMC Utrecht, Utrecht, the Netherlands
L.L.J. Schouten, Maastricht University, Maastricht, the Netherlands
J.H. Moore, Geisel School of Medicine, Dartmouth College, Lebanon, USA
N.C. Onland-Moret, Julius Center, UMC Utrecht, Utrecht, the Netherlands
Genome-wide association studies (GWAS) have identified 19 loci associated with ovarian cancer risk. However, these single nucleotide polymorphisms (SNPs) explain only a small fraction of ovarian cancer susceptibility. Possibly, because these one-SNP-at-a-time analyses ignore the complexity of the underlying biological mechanisms. Here, we aim to embrace this complexity by detecting and characterizing higher-order SNP-SNP interactions that are associated with ovarian cancer susceptibility.
We used an integrated approach to search for higher-order interactions between 968 SNPs in association with ovarian cancer risk. Within EPIC we designed a nested case-control study including 590 ovarian cancer cases and 1190 controls. First, we assessed all possible second order interaction using information gain measures. Next, a statistical interaction network
was constructed based on the strongest pairwise interactions. Based on this visualization, we prioritized SNPs for an exhaustive search of all 2nd to 4th order interactions with multifactor dimensionality reduction (MDR). Finally, we used logistic regression analyses to quantify our findings.
Multiple statistically significant higher-order interaction models were detected in our study. The model with the highest odds ratio (OR) represented an interaction between rs3003917 in ESR1, rs3217907 in CCND2, and rs2282411 in CCNA1 (OR: 2.18, 95% CI: 1.72-2.77).
By combining multiple machine learning methods, we were able to identify and characterize SNP combinations that were associated with an increased risk of ovarian cancer. Based on these results we can generate new hypotheses about the etiology of ovarian cancer and guide future genetic association.
Published: 06 Jun, 2013