Evaluation of the performance of two prediction models, the IrisPlex and the novel EC12 model, for eye colour predictions in a Norwegian population.

Abstract

Biological material obtained from a crime scene is used to generate DNA-profile by typing short tandem repeat (STR) markers. However, if the STR-profile do not match the DNA profile of suspects or from a crime DNA database, the investigation can go towards typing markers that can estimate externally visible characteristics (EVCs). EVCs can function as a “biological witness” and thus aid a police investigation.

In this work the IrisPlex prediction model and a novel prediction model, EC12, were evaluated in 521 samples from the Norwegian population. A PCR-SBE-CE assay amplifying the fourteen SNPs included in the two models was optimised at Section of Forensic Genetics, Copenhagen, Denmark before it was established at Centre of Forensic Genetics, Tromsø, Norway.

IrisPlex showed high prediction accuracy for blue and brown eye colour (AUC-value of 0.84 and 0.94, respectively). However, the model did not perform good in prediction of intermediate eye colour (AUC-value of 0.6), which represented 24% of the Norwegian population and thus all these individuals were incorrectly predicted.

Comparison of EC12 and an adjusted IrisPlex model (IP NO) showed a small increase in correct predictions from 72% to 75%, respectively. A higher prediction accuracy for all eye colours were observed for the EC12 model, with AUC-value of 0.84 (blue), 0.97 (brown) and 0.68 (intermediate), while IP NO obtained AUC-values of 0.81 (blue), 0.93 (brown) and 0.59 (intermediate). This increase may imply that the additionally SNPs included in this model has an improving effect on eye colour prediction. However, the prediction of intermediate eye colour was still not good, indicating the importance of further phenotypic investigation of this category.