What Genes Tell about Iris Appearance

摘要

Predicting phenotypes based on genotypes is generally hard, but has shown good results for prediction of iris color. We propose to correlate the appearance of iris with DNA. Six single-nucleotide polymorphisms (SNPs) have previously been shown to correlate with human iris color, and we demonstrate that especially one of the six SNPs are correlated with iris appearance. To perform this analysis we need a method to model the iris appearance, and we suggest an iris characterization based on a bag of visual words, which gives us a similarity measure between images of eyes. We have a dataset of 215 eye images with corresponding SNP types, where the image of the iris has been segmented. We perform two experiments based on the iris characterization. An agglomerative clustering is performed and the result is that one SNP - rsl2913832 (HERC2) is highly correlated with the image clustering. Furthermore subspace projections are performed supporting that this SNP is very important for eye color expression. With the suggested image characterizations we are able to investigate the correlation between the phenotypic iris appearance and specific SNPs. This has potential for further investigation of the relation between DNA and iris appearance, especially with focus on iris texture.