f₄-statistics-based ancestry profiling and convolutional neural network phenotyping shed new light on the structure of genetic and spike shape diversity in Aegilops tauschii Coss.

Abstract

Aegilops tauschii Coss., a progenitor of bread wheat, is an important wild genetic resource for breeding. The species comprises three genetically defined lineages (TauL1, TauL2, and TauL3), each displaying distinctive phenotypes in various agronomic traits, including spike shape. In the present work, we studied the relationship between population structure and spike shape variation patterns using a collection of 249 accessions. f₄-statistics-based ancestry profiling confirmed the previously identified lineages and revealed a genetic component derived from TauL3 in the genomes of some southern Caspian and Transcaucasus TauL1 and TauL2 accessions. Spike shape variation patterns were analyzed using a convolutional neural network-based approach, trained on green and dry spike image datasets. This analysis showed that spike shape diversity is structured according to lineages and demonstrated that the lineages can be distinguished based on spike shape. The implications of these findings for the origins of common wheat and the intraspecific taxonomy of Ae. tauschii are discussed.

Tracing origins through spike shape using machine learning

Wild wheat, Aegilops tauschii, harbors a vast reservoir of alleles that remain untapped in breeding programs. These alleles could contribute valuable traits, such as drought tolerance and disease resistance, when introduced into bread wheat. To fully harness this genetic diversity, it is crucial to rapidly identify strains carrying potentially beneficially alleles. In Ae. tauschii, which is composed of strain groups (lineages) with unique genetic makeups, this can be done by determining a strain’s lineage based on spike shape. In this work, we trained machine learning models to predict lineage based on spike morphology and found that spike shape diversity mirrors lineage structure. These models demonstrated potential for practical use in assigning strains to their respective lineages based on spike shape. This work opens new avenues for the application of machine learning in wheat improvement, as well as in the genetic and evolutionary studies of wheat morphology.