From Images to Loci: Applying 3D Deep Learning to Enable Multivariate and Multitemporal Digital Phenotyping and...

Area of Science:

• Agricultural Science
• Genetics
• Plant Science

Background:

• Nitrogen use efficiency (NUE) in wheat is crucial for sustainable agriculture, reducing fertilizer needs while maintaining yield and quality.
• Locating NUE genes requires extensive phenotypic data, which is often a bottleneck in breeding programs.
• Traditional phenotyping methods can be labor-intensive and may not capture dynamic plant growth effectively.

Purpose of the Study:

• To develop and validate a 3D digital dynamic phenotyping approach for wheat.
• To identify novel genetic loci associated with nitrogen use efficiency (NUE) and plant height in wheat.
• To provide a scalable method for generating high-throughput phenotypic data for wheat breeding.

Main Methods:

• Utilized low-altitude aerial photography to capture large-scale field images, generating 3D point clouds and multispectral data.
• Developed a wheat 3D plot segmentation dataset and quantified plot canopy height using PointNet++.
• Extracted dynamic digital phenotypes (height- and vegetation index-related) from time-series data and applied them to genome-wide association studies (GWAS).

Main Results:

• Successfully generated dynamic digital phenotypes, including height and nitrogen utilization-related vegetation indices.
• Applied these phenotypes to GWAS in 160 wheat cultivars, identifying reliable loci associated with plant height and NUE.
• Some identified loci were consistent with previously published findings, validating the approach.

Conclusions:

• 3D digital dynamic phenotyping offers a valuable advancement for efficiently locating NUE genes in wheat.
• The developed methodology provides accurate phenotypic data essential for breeding nitrogen-efficient wheat varieties.
• This approach supports the sustainable development of agriculture by facilitating the selection of improved wheat cultivars.