GenoEye: A machine learning-based framework for the prediction of intermediate eye color phenotypes.

Area of Science:

• Forensic genetics
• Computational biology
• Human genetics

Background:

• Forensic DNA phenotyping (FDP) predicts externally visible characteristics from DNA.
• Accurate prediction of intermediate eye colors (green, hazel) is challenging, especially in specific populations.
• Existing FDP tools have limited sensitivity for intermediate eye color phenotypes.

Purpose of the Study:

• To develop an interpretable machine learning framework (GenoEye) for three-category eye color prediction.
• To improve the prediction accuracy of intermediate eye colors using an expanded SNP panel.
• To provide a robust and interpretable tool for forensic applications.

Main Methods:

• Developed GenoEye, a gradient boosting machine learning framework.
• Utilized an expanded single nucleotide polymorphism (SNP) panel, deriving a 37-SNP predictive signature.
• Validated the model on a cohort of 363 Italian individuals and independent data.

Main Results:

• GenoEye achieved high accuracy for blue and brown eye color prediction (AUC up to 0.97).
• Significantly improved classification of intermediate eye colors (AUC = 0.79) compared to existing methods.
• Demonstrated higher sensitivity for intermediate eye colors while maintaining high specificity.

Conclusions:

• GenoEye enhances the resolution of intermediate eye color prediction in forensic DNA phenotyping.
• The framework offers a robust, interpretable, and web-based tool for practical forensic use.
• This advancement aids investigations by providing more detailed phenotypic information from genetic data.