A few years ago there was a paper out which illustrated that standard Galtonian method of regression of offspring upon parents still predicted height far better than the most modern genomic techniques. The issue is that height is a quantitative trait whose variation is controlled by variants at hundreds, and likely thousands, of loci. Generating a useful prediction for one individual from a "bottom-up" genetic model is daunting because of the overwhelming number of variants. This is in contrast to pigmentation traits, which been found to be well characterized by a few large effect quantitative trait loci. That is, one gene can account for a substantial minority of the variation within the population of the trait. In regards to eye pigmentation in Europeans the majority of the blue vs. non-blue eye color difference can be accounted for by one locus, HERC2-OCA2. Not so for height, intelligence, and now it seems likely, facial morphology. A Genome-Wide Association Study Identifies Five Loci Influencing Facial Morphology in Europeans. The issue I'm alluding to is found buried in the paper:
Moreover, our data also highlight that the high heritability of facial shape phenotypes (as estimated here and elsewhere), similar to that of adult body height...involves many common DNA variants with relatively small phenotypic effects. Future GWAS on the facial phenotype should therefore employ increased sample sizes as this has helped to identify more genes for many other complex human phenotypes such as height...and various human diseases.
Am I on crack, or are we not going to really get much yield from hunting for specific genes in most cases? Granted, there may be populations like pygmies who are well deviated on some trait, so we'll put a finger on particular variants which shift the trait value a lot. Additionally, like height I wouldn't be surprised if we find some evidence for selection on facial morphology related genes. But this seems to dampen the likelihood of robust individual prediction. The limits of forensic genetics? Say it ain't so!