Explainable artificial intelligence for differentiating honey bee genotypes using morphometrics and SSR markers

Abstract

This study aims to classify honey bee genotypes by integrating explainable artificial intelligence techniques, particularly decision trees, with both morphometric and molecular data. A total of 4949 samples were collected from 500 colonies across five regions in Türkiye, representing diverse subspecies and ecotypes. Morphometric data included 16 key wing characteristics, while molecular data contained 26 highly informative SSR loci. First, we used 16 morphometric wing parameters to classify bees into five regions where they originate. The decision tree algorithm resulted in a tree with wing length and O26 and L13 angles, but the classification accuracy was low (51%). Later, we included 26 molecular variables and obtained a decision tree with four SSR loci—Ap218, Ap274, Ap001, and Ap289—and achieved a high classification accuracy of 96.38%. The findings also revealed the first-ever identification of a SSR locus (Ap218) strongly associated with wing length in honey bees. Finally, we explained wing length with molecular data by modeling a regression decision tree. This tree identified Ap218, Ap223, and Ap001 as the most significant SSR loci for the wing length model. This study provides a powerful approach for differentiating honey bee genotypes while offering valuable insights into the genetic factors influencing wing morphology. The results have significant implications for the conservation and sustainable management of honey bee genetic resources, particularly in regions like Türkiye where genetic diversity is at risk.