Objective To evaluate the accuracy of using HIrisPlex-S system for eye color, hair color, and skin color prediction in forensic DNA phenotyping (FDP).

Methods A total of 49 unrelated individuals were enrolled. First, based on the collected facial photographs and questionnaire responses, the actual phenotypic traits, including eye color, hair color, and skin color, of the participants were manually assessed and categorized. Subsequently, peripheral blood samples were collected from the participants, and DNA was extracted. Genotyping of the HIrisPlex-S system was performed using the SNaPshot method combined with capillary electrophoresis. Single nucleotide polymorphism (SNP) sites were analyzed using the online HIrisPlex-S prediction platform (https://HIrisPlex.erasmusmc.nl/) to predict the aforementioned phenotypic traits. Two different methods were used to interpret the prediction results. In Method 1, the phenotypic trait with the highest probability value were directly selected as the predicted phenotype. In Method 2, the prediction guidelines proposed by Manfred Kayser's team was applied. The sensitivity and specificity of the predictions were calculated by comparing the prediction results with the manually interpreted actual phenotypic data, and the differences between the two interpretation methods in these performance metrics were compared. Additionally, two real case samples were included for phenotypic inference analysis to validate the system's effectiveness in practical applications.

Results Among the participants, approximately 55% had "brown" eyes, while "intermediate" eye color was less common; approximately 45% had "black" hair; and approximately 39% had "pale" skin. After optimizing the concentration of single base extension (SBE) primers, the peak balance of most SNP loci in the HIrisPlex-S system improved, though inter-laboratory reproducibility remained suboptimal. In eye color prediction, the sensitivity of Method 1 (0.8750, 0.0000, and 1.0000) was slightly higher than that of Method 2 (0.8125, 0.0000, and 0.9259), while the specificity of Method 1 (0.8485, 1.0000, and 0.8182) was slightly lower than that of Method 2 (0.9091, 1.0000, and 0.9091). For hair color prediction, the two methods showed identical sensitivity (0.7727, 0.6154, and 0.8571), but Method 2 showed slightly better specificity (1.0000, 0.9444, and 0.7142) than Method 1 (1.0000, 0.9167, and 0.7429). In skin color prediction, both the sensitivity (0.0000, 0.7895, 1.0000, 0.3333, and 0.8750) and specificity (1.0000, 0.9667, 0.8333, 0.9767, and 0.8780) of Method 2 were slightly higher than those of Method 1 (sensitivity: 0.0000, 0.5789, 0.9231, 0.1667, and 0.7500; specificity: 1.0000, 0.9667, 0.6944, 0.9535, and 0.8780). However, overall, there was no statistically significant difference in sensitivity or specificity between the two methods (P > 0.05). Additionally, the phenotypic predictions for the two real case samples were consistent with genetic ancestry analysis.

Conclusion The HIrisPlex-S system can be used to predict the eye and hair colors in population samples, but its accuracy in skin color prediction needs further improvement.