Genetics. 2025 Sep 05. pii: iyaf184. [Epub ahead of print]
Phantom epistasis arises when, in the course of testing for gene-by-gene interactions, the omission of a causal variant with a purely additive effect on the phenotype causes the spurious inference of a significant interaction between two SNPs. This is more likely to arise when the two SNPs are in relatively close proximity, so while true epistasis between nearby variants could be commonplace, in practice there is no reliable way of telling apart true epistatic signals from false positives. By considering the causes of phantom epistasis from a genealogy-based perspective, we leverage the rich information contained within reconstructed genealogies (in the form of ancestral recombination graphs) to address this problem. We propose a novel method for explicitly quantifying the genealogical evidence that a given pairwise interaction is the result of phantom epistasis, which can be applied to pairs of SNPs regardless of the genetic distance between them. Our method uses only publicly available data and so does not require access to the phenotypes and genotypes used for detecting interactions. Using simulations, we show that the method has excellent performance at even low genetic distances (around 0.5cM), and demonstrate its power to detect phantom epistasis using real data from previous studies. This opens up the exciting possibility of distinguishing spurious interactions in cis from those reflecting real biological effects.
Keywords: ancestral recombination graphs; gene-by-gene interactions; genome-wide association studies; phantom epistasis