A1416
Title: DNA-based predictive models for human appearance characteristics in forensic science
Authors: Joanna Karlowska-Pik - Nicolaus Copernicus University in Torun (Poland) [presenting]
Abstract: Genetic predictive models of human appearance aid forensic identification or narrow suspect pools. Complex human traits are highly polygenic, with individual variants exerting small effects and often pleiotropic influences across phenotypes. Analysis of 240 candidate SNPs (single-nucleotide polymorphisms) in 999 Polish individuals employed multiple machine learning methods for missing-data imputation, feature selection, and predictive modelling, complemented by traditional statistical approaches. SNP-SNP interactions were evaluated using a nonparametric, entropy-based mutual information interaction. The analysis detected, for example, two novel association signals displaying broad pleiotropic associations with hair-related traits and participating in several interaction effects. Findings illuminate the genetic overlap and epistatic architecture of human appearance phenotypes and develop DNA-based predictive models for forensic phenotyping.
