Journal of Forensic Research

Prediction Model Validation: Normal Human Pigmentation Variation

Abstract

Robert K. Valenzuela, Shosuke Ito, Kazumasa Wakamatsu and Murray H. Brilliant

In a past study, we developed multiple linear regression (MLR) models that employed three single nucleotide polymorphisms (SNPs) that predicted a significant proportion of variation in pigmentation phenotypes from a large population cohort (n=789, training sample). Multiple linear regression models were developed for skin reflectance, eye color, and two aspects of hair color (log of the ratio of eumelanin-to-pheomelanin and total melanin). In this report, using an independent cohort (n=242 , test sample), we 1) externally cross-validated the prediction models, and 2) tested and refined the algorithm presented in the study by Valenzuela and colleagues, (2010). Relative shrinkage was moderate for skin reflectance (23.4%), eye color (19.4%), and the log of the ratio of eumelanin-to-pheomelanin in hair (37.3%), and largest for total melanin (67%) in hair. Independent construction of predictive models using our algorithm for the test sample set yielded the same or similar models as the training sample set. Two of the three SNPs composing the models were the same, with some variability in the third SNP of the model.