Investigating the Genetic Architecture of Stomatal Density in Wheat using a Convolutional Neural Network and GWAS

Abstract

none

Bread wheat (Triticum aestivum) is one of the most important crops in the world and is grown in a wide range of environments. Wheat yields are mainly limited by the lack of precipitations in certain parts of the world. Stomatal density (SD) and stomatal area (SA) are important traits for gas exchanges between the plant and the atmosphere, and thus, for water-use efficiency. Therefore, understanding the genetic architecture of SD and SA is essential for the breeding of drought tolerant wheat. However, the ability to discover genetic loci controlling stomatal traits has been hindered by the low throughput manual phenotyping methods employed for measuring SD and SA. We used a deep learning method to automatically measure SD and SA on 133 bread wheat accessions. The automatic measurements of SD were compared with SD measured manually. The deep learning model was able to accurately detect stomata with a precision, recall and F1-score of 0.990, 0.982 and 0.986 respectively. A genome-wide association study (GWAS) identified 58 quantitative trait loci associated with SA as well as automatically and manually measured SD. QTL were consistently detected between manual and automatic phenotyping. Thus, the two methods can be used in conjunction in order to validate the detected loci. Our results demonstrate that deep learning can be used to investigate the diversity and genetic control of SD on large populations accurately.