Searching for improvements in predicting human eye colour from DNA.

Area of Science:

• Genetics
• Bioinformatics

Background:

• Human genome variability offers predictive potential for physical traits.
• Predicting pigmentation phenotypes like eye color remains challenging.
• Advancements in DNA analysis are crucial for forensic and anthropological applications.

Purpose of the Study:

• To discover novel single nucleotide polymorphism (SNP) markers for human eye color prediction.
• To enhance the accuracy of eye color prediction models using advanced algorithms.
• To evaluate the efficacy of machine learning in predicting complex phenotypes.

Main Methods:

• Whole-exome sequencing of 150 Polish individuals to identify candidate SNPs.
• Quantitative eye color phenotyping using high-resolution iris imaging and DIAT software.
• Predictive modeling on an independent set of 849 samples using selected SNPs and machine learning algorithms (e.g., LASSO, HyperLasso).

Main Results:

• Identified 27 novel candidate SNP markers associated with human eye color.
• Achieved high prediction accuracies using LASSO-regularized and BIC-based regression models.
• A new variant, rs2253104 in the ARFIP2 gene, demonstrated predictive potential.
• Machine learning significantly increased intermediate eye color prediction sensitivity by up to 39% compared to the IrisPlex model.

Conclusions:

• Discovered new potential genetic predictors for human eye color.
• Demonstrated the effectiveness of advanced machine learning algorithms in improving eye color prediction.
• The findings offer valuable insights for developing future predictive models in forensic and anthropological genetics.