APTES: a high-throughput deep learning-based Arabidopsis phenotypic trait estimation system for individual leaves and...

Area of Science:

• Plant Science
• Computational Biology
• Genetics

Background:

• High-throughput phenotyping is crucial for understanding plant growth and organ development in model organisms like Arabidopsis thaliana.
• Existing methods for trait estimation can be time-consuming and lack precision, hindering large-scale studies.

Purpose of the Study:

• To develop an open-access, automated system for rapid and precise estimation of leaf and silique traits in Arabidopsis thaliana.
• To leverage computer vision and deep learning for enhanced plant phenotyping accuracy and throughput.

Main Methods:

• Developed the Arabidopsis Phenotypic Trait Estimation System (APTES), utilizing enhanced Mask R-CNN for leaf segmentation and DetectoRS for silique segmentation.
• Employed deep learning models to extract 64 leaf and 64 silique traits from plant photographs.
• Performed genome-wide association studies (GWAS) on phenotyped Arabidopsis accessions to identify trait-associated SNPs.

Main Results:

• APTES achieved high segmentation accuracy for leaves (F1 score 0.961) and siliques (F1 score 0.942), outperforming baseline models.
• Automated trait parameter calculation showed strong correlation (R² 0.776–0.976) and low error rates (1.89%–7.90% MAPE).
• GWAS identified 1,042 significant SNPs associated with 18 traits and one SNP linked to silique number.

Conclusions:

• APTES provides a valuable, automated solution for leaf and silique segmentation and trait estimation in Arabidopsis thaliana.
• The system's validated performance across diverse datasets and species highlights its broad applicability in plant science.
• APTES facilitates high-throughput phenotyping, accelerating genetic research and crop improvement efforts.