[bordercases]

Not with any reasonable accuracy for most traits - yet. There are both many to one and one to one relationships for genes to phenotype. Add in continuous/fuzzy/probabilistic measures and you get an explosion for the total set (although some subcollection might be tractable with data.)

Observationally some traits are more straightforward. Skin color, IQ and and height give us an unweighted average as the expected value for the offspring, with the height of either parent setting bounds on the variance. But that isn't news if your service can be replaced by SAT scores and simply taking a look in the mirror.

It's a good idea though.