Integrative machine learning approach for identifying genes associated with quantitative traits: A soybean (Glycine...

Area of Science:

• Genomics
• Plant Breeding
• Artificial Intelligence

Background:

• Traditional methods for identifying molecular markers for quantitative traits are inefficient.
• Minor-effect markers have limited impact in practical breeding programs.
• Improving soybean yield requires advanced genomic strategies.

Purpose of the Study:

• To enhance the identification of minor-effect molecular markers and genes linked to quantitative traits.
• To address inefficiencies in traditional marker mining and improve marker impact in breeding.
• To leverage AI for genomic analysis in soybean to boost yield.

Main Methods:

• Analysis of over 15,000 soybean genotypes across nine maturity groups.
• Application of an AI-driven genomic approach, including the Light Gradient Boosting Machine (LightGBM) model.
• Validation using fivefold cross-validation, t-tests, and assessment of genotype-by-environment interactions.

Main Results:

• Discovery of 2513 key genetic markers and 393 genes associated with soybean yield.
• LightGBM model achieved high accuracy (R²=0.8369) in predicting yield.
• Identified both stable and environment-dependent genetic determinants for yield, with region-specific marker distributions.

Conclusions:

• The AI-driven LightGBM approach is reliable and robust for soybean yield prediction.
• Introduced three principles for marker-assisted selection (MAS): marker complementarity, regional specificity, and quantification.
• The findings offer refined targets for precision breeding and insights into genotype-by-environment interactions for enhanced agricultural productivity.