epistasis

Huang W & Mackay TFC 2016 The genetic architecture of quantitative traits cannot be inferred from variance component analysis. PLoS Genet 12:e1006421.

• ideally, a variance component partition should have a one-to-one corresponding relationship with gene actions in order for it to measure the relative importance of gene actions
• the classical V_A + V_D + V_I partition obviously does not possess this property, despite it being an orthogonal partition (uncorrelated variance components) and having suggestive names
• except for additive gene actions, which contribute only to V_A, both dominant and epistatic gene actions contribute to multiple variance components
• it is clear that this classical V_A + V_D + V_I partition is a poor indicator of the underlying genetic architecture
• purely epistatic genetic architecture can often result in a partition where V_A is large but V_I is small

• the concept of additive variance does not carry with it the assumption of additive gene action
• the existence of additive variance is not an indication that any of the genes act additively

• perhaps the best way to counter the argument that large V_A is evidence for unimportance of non-additive gene actions is to derive alternative ways of partitioning variance where one of the non-additive components dominates others, a property that has been shown previously only for V_A
• this turns out to be easy if the non-additive components are given the priority to explain the genetic variation, as does V_A in the classical model