Testing the Impact of Trait Prevalence Priors in Bayesian-Based Genetic Prediction Modeling of Human Appearance Traits

The study investigated the impact of using trait prevalence-informed priors in Bayesian models for predicting human appearance traits such as eye, hair, and skin color, as well as hair structure and freckles. It found that while these priors could potentially improve prediction accuracy, their effectiveness varied across different traits and categories. Importantly, incorrect specification of these priors often reduced accuracy compared to models without priors. The study highlighted the challenge posed by the limited knowledge of spatial prevalence for appearance traits, making the use of prevalence-informed priors in genetic prediction models currently impractical.