Computer vision and machine learning for robust phenotyping in genome-wide studies.

Area of Science:

• Agricultural Science
• Plant Genetics
• Computational Biology

Background:

• Traditional crop stress evaluation is slow and labor-intensive, hindering genetic trait analysis.
• Iron deficiency chlorosis (IDC) is a significant abiotic stress affecting soybean yield.
• Developing efficient phenotyping methods is crucial for genetic improvement.

Purpose of the Study:

• To develop and validate a machine learning-enabled image-phenotyping pipeline for soybean IDC.
• To integrate automated phenotyping with genome-wide association studies (GWAS) and genomic prediction.
• To accelerate the genetic dissection of quantitative traits related to abiotic stress tolerance.

Main Methods:

• An end-to-end phenotyping workflow was designed, including optimized image capture and canopy identification.
• Domain-expert informed features were extracted from images to quantify IDC expression.
• Supervised machine learning classifiers were trained to link image features with IDC scores.
• ML-generated phenotypic data were used for GWAS and genomic prediction in a soybean association panel.

Main Results:

• The ML pipeline accurately classified and scored IDC severity across 461 soybean accessions.
• Genome-wide association studies identified a previously known locus and a novel locus associated with IDC.
• The novel locus contains a gene homolog involved in iron acquisition, highlighting its role in IDC tolerance.
• The study demonstrated the reliability and advantages of the ML-enabled image-phenotyping pipeline.

Conclusions:

• The developed ML pipeline offers a robust and rapid phenotyping framework for ground-based systems.
• This approach significantly enhances the efficiency of genetic studies for abiotic stress tolerance in soybean.
• Integrating automated phenotyping with genomic prediction provides a powerful tool for crop improvement.