Using Machine Learning to Develop a Fully Automated Soybean Nodule Acquisition Pipeline (SNAP)

Area of Science:

• Agricultural Science
• Plant Biology
• Biotechnology

Background:

• Soybean root nodules, formed by Bradyrhizobium japonicum symbiosis, are crucial for fixing atmospheric nitrogen into ammonia for plant growth.
• Current nodule quantification methods are labor-intensive, subjective, and lack detailed information, hindering efficient phenotyping and breeding.

Purpose of the Study:

• To develop and validate an automated Soybean Nodule Acquisition Pipeline (SNAP) for accurate and high-throughput nodule quantification on soybean roots.
• To overcome the limitations of manual counting and subjective assessments in nodule analysis.

Main Methods:

• Developed SNAP by integrating RetinaNet and UNet deep learning models for nodule detection and segmentation.
• Trained and validated SNAP using a diverse dataset of 691 soybean roots across various genotypes, growth stages, and locations.
• Achieved a high model fit with R² = 0.99.

Main Results:

• SNAP significantly reduces labor and inconsistencies associated with manual nodule counting.
• The pipeline provides quantifiable data on nodule number, growth, location, and distribution.
• High throughput phenotyping enables early-stage assessment of genetic and environmental factors influencing nodulation.

Conclusions:

• SNAP offers a robust and efficient solution for soybean nodule quantification, enhancing research capabilities.
• The technology facilitates the assessment of factors affecting nodulation, potentially leading to improved nitrogen use efficiency in soybean and other legumes.
• SNAP advances the understanding of plant-Bradyrhizobium interactions and supports breeding programs for enhanced crop performance.