Evaluating 10 Pigmentation Associated SNPs for Forensic DNA Phenotyping in Saudi Arabians: Population Structure and Model Calibration

Forensic DNA phenotyping (FDP) predicts externally visible characteristics from biological evidence, providing crucial investigative leads when conventional STR profiling fails. Although the HIrisPlex-S system is well-validated in European populations, its performance in Middle Eastern groups remains understudied. This study evaluated the utility of 10 pigmentation-associated SNPs (selected from the HIrisPlex-S panel) in a Saudi Arabian cohort through comprehensive population genetic analyses, quality control metrics, and machine learning approaches. Among 100 unrelated Saudis (50 males, 50 females; mean age 24.4 +/- 10.3 years), four SNPs were monomorphic, reducing standard panel informativeness. Exact Hardy-Weinberg equilibrium testing revealed significant deviation in four informative SNPs (p < 0.05). Saudis showed closest genetic affinity to South Asians (FST = 0.030) and moderate European differentiation (FST = 0.060). HERC2 rs12913832 predicted brown versus intermediate eye color with 88% accuracy (ordinal-score AUC = 0.66, 95% CI: 0.53–0.80); TYRP1 rs683 alone achieved 73% accuracy (AUC = 0.68) for black hair prediction in this cohort. rs2402130 (SLC24A4) and rs1042602 (TYR) did not show statistically significant associations with hair color (chi2 = 5.116, p = 0.078 and chi2 = 4.683, p = 0.096, respectively); a combined model including rs683 with either non-significant SNP did not materially improve predictive performance beyond rs683 alone. For hair colour, an exploratory custom logistic regression model using rs683 achieved 73% accuracy (AUC = 0.68, 95% CI: 0.58-0.78) for black versus non-black classification in internal cross-validation, modestly exceeding the no-information majority class baseline (68% accuracy, AUC = 0.50). No valid comparison with the complete HIrisPlex-S hair model was possible because the 10-SNP subset lacked 12 of 22 HIrisPlex hair predictors, including all 11 MC1R variants. For eye color, the custom single-SNP model and the partial HIrisPlex-S profile performed comparably (AUC = 0.66 vs. 0.63). Saudi-specific allele frequencies and high monomorphism at key MC1R/OCA2 variants limit standard FDP panel effectiveness, underscoring the necessity for population-tailored prediction frameworks. Notably, the absence of blue eyed and red/blond haired individuals in this cohort precluded validation of predictions for lighter phenotypes; model evaluation was therefore restricted to brown versus intermediate eye color and black versus non-black hair classifications.