Generalized Linear Model with Elastic Net Regularization and Convolutional Neural Network for Evaluating Aphanomyces...

Area of Science:

• Plant pathology
• Agricultural science
• Computational biology

Background:

• Phenomics enables large-scale, quantitative plant trait assessment.
• Advanced data analysis is crucial for leveraging phenomics data.
• Aphanomyces root rot (ARR) is a significant disease affecting lentil production.

Purpose of the Study:

• To evaluate Aphanomyces root rot (ARR) resistance in lentil accessions using image-based phenotyping.
• To develop and compare machine learning models for classifying ARR resistance.
• To investigate the effectiveness of phenomics and machine learning in quantifying plant disease resistance.

Main Methods:

• Utilized Red-Green-Blue (RGB) images of lentil roots from 547 accessions.
• Created a dataset of 6,460 annotated root images based on disease severity.
• Developed and applied two classification approaches: generalized linear model with elastic net regularization (EN) and convolutional neural network (CNN).

Main Results:

• The EN model achieved a classification accuracy of 0.91 (±0.004), outperforming the CNN's accuracy of 0.84 (±0.009).
• Both methods accurately identified the resistant class, while the partially resistant class presented classification challenges due to overlapping features.
• EN model demonstrated superior performance in classifying resistant (0.96 ± 0.005), partially resistant (0.82 ± 0.009), and susceptible (0.92 ± 0.007) categories.

Conclusions:

• Phenomics technologies combined with machine learning provide effective quantitative measures for ARR resistance in lentil.
• The generalized linear model with elastic net regularization (EN) is a promising approach for high-throughput phenotyping of plant disease resistance.
• Further research is needed to improve classification accuracy for intermediate resistance levels.