Abstract for presentation at 11th International Congress of Human Genetics

Most human diseases result from complex interactions among multiple genes that yield weak or modest effects. Despite the growing awareness of the importance of gene–gene interactions, the paradigm of detectable effect of individual variants remains the cornerstone of genome association studies with tagSNPs. The interactive effect of two variants is only tested once the individual effect of one variant is detected. Both genes however, may have at the same time a weak (or even no) marginal effect and an important effect through their interaction. In such a situation, current approaches may fail to detect variants having a crucial role in the causal chain.
We propose here an alternative strategy that allows the detection of the involvement of two genes without individual effect. This strategy simultaneously uses information on candidate polymorphisms in two genes A and B. We first estimate the relative marginal penetrances of the genotype at each locus [F_A] and [F_B] and of the joint genotype [F_AB]. The null hypothesis of no interaction corresponds to the following equality: [F_AB] = [F_A]^T x [F_B] which may be tested by a chi-square with four degrees of freedom. Under the alternative hypothesis of interaction, the statistic follows a non-central chi-square distribution. The non-centrality parameter provides the power of the test. We show that our approach has a good power to detect the effect of two genes in situations for which locus-by-locus test would have been unsuccessful. At a time where genome-wide association studies are fashionable, we think it is important to consider the alternative strategy of studying good candidate pathways with our approach. This requires preliminary biological studies to define relevant pathways and variants.