Abstract
BACKGROUND: The external human ear is a polymorphic and polygenic structure with individual uniqueness, making it a valuable target in forensic DNA phenotyping (FDP) studies. Previous genome-wide association studies have identified multiple genetic loci associated with variation in ear characteristics. However, research focused on predicting ear morphology within the context of FDP remains limited. This study aimed to develop DNA-based predictive models for external ear morphology in the Chinese population. METHODS: Digital photographs of 675 volunteers were used to score 13 ear phenotypes, each categorized into three levels. Multinomial logistic regression (MLR) was applied for genetic association analysis. Five predictive models-MLR, support vector machines, random forest, AdaBoost, and k-nearest neighbors-were developed and evaluated using 10-fold cross-validation. RESULTS: Genetic association analysis identified several influential single-nucleotide polymorphism (SNPs) for each ear phenotype. Among the five models, AdaBoost and MLR demonstrated superior performance, achieving area under the curve (AUC) values above 0.7 for predicting absent tragus cases (level_0). To simplify classification, binary models incorporating genetic interactions were constructed for absent tragus cases. Specifically, the AdaBoost model achieved an AUC of 0.74, while the binary logistic regression (BLR) model reached an AUC of 0.72. CONCLUSIONS: These findings highlight the potential forensic application of genetic markers in predicting ear morphology within the Chinese population, contributing to the advancement of FDP research and practice.