Integrating machine learning and the GGE biplot for identification of climate-suitable grasspea genotypes

Area of Science:

• Agricultural Science
• Plant Breeding
• Genetics

Background:

• Grasspea (Lathyrus sativus) is a resilient, nutrient-rich legume crop vital for challenging agro-ecosystems.
• Limited information exists on grasspea genotype recommendations for general and specific adaptations.
• Understanding genotype-environment interactions is crucial for optimizing crop performance.

Purpose of the Study:

• To identify stable grasspea genotypes by analyzing genotype-environment interactions.
• To pinpoint suitable locations in India for future grasspea evaluations.
• To validate and predict genotype performance using machine learning algorithms.

Main Methods:

• A panel of 64 diverse grasspea genotypes was evaluated across four Indian locations using the GGE biplot approach.
• Multi-trait performance was assessed to calculate selection indices for ranking genotypes and environments.
• Machine learning models, including Random Forest (RF) and Multilayer Perceptron (MLP), were used for validation and prediction.

Main Results:

• The environment significantly influenced trait variation, followed by genotype × environment interactions.
• Several genotypes, including FLRP-B54-1-S2, Prateek, 31-GP-F3-S7, and 48-GP-F3-S3, demonstrated excellent multi-trait performance.
• The Random Forest model exhibited superior predictive accuracy (R² values up to 0.947) compared to the MLP model.

Conclusions:

• Genotypes 'Prateek,' '31-GP-F3-S7,' and '48-GP-F3-S3' are the most stable based on combined yield and trait performance.
• These stable genotypes are recommended for commercial cultivation in areas prone to weather extremes.
• The study provides valuable insights for grasspea breeding programs and regional adaptation strategies.