Causal Meaning of Genomic Predictors: Implication on Genome-Enabled Selection Modeling

Abstract Text: The term “effect” in additive genetic effect suggests a causal meaning. However, inferences on such quantities for selection purposes are normally conducted as prediction tasks. Predictive ability is currently the most used criterion for comparing models and evaluating new methodologies, but it is insufficient to evaluate if predictors identify causal effects. Therefore, the usual approach to infer genetic effects seems to contradict the label of the quantity inferred. Here we investigate if genomic predictors for selection should be treated as standard predictors from regression models, or if they must reflect a causal effect, asking for causal inference approaches. We demonstrate that selection requires learning causal genetic effects. However, genomic predictors may reflect non-causal signal, providing good predictions but poorly representing true genetic effects. Genomic selection models should be constructed aiming primarily for identifiability of causal genetic effects, not for predictive ability.

Keywords:

causal inference

genomic selection

model comparison

prediction