Yes, it's been a painful lesson for me to learn: you can often build a good-enough approximation of the underlying physics to make good-enough predictions, even

Yes, it's been a painful lesson for me to learn: you can often build a good-enough approximation of the underlying physics to make good-enough predictions, even without modelling all the molecular details directly, as long as you have enough data, good algorithms, and fast computers. https://www.annualreviews.org/doi/full/10.1146/annurev-genom-120121-102607 https://www.annualreviews.org/doi/full/10.1146/annurev-genom... is a review of research in this area. I don't think it's particularly controversial because realistically, the underlying data supports the hypothesis that facial features are hereditable and then do association studies to find plausible candidate multivariate genomic features that predict them accurately. I think that's far enough from controversial "Race science" that it's hard for people to make reasonable criticisms of this research.