Toward a general framework for AI-enabled prediction in crop improvement

Area of Science:

• Agricultural Science
• Computational Biology
• Genetics

Background:

• Genomic prediction for complex traits faces challenges due to the curse of dimensionality.
• Existing methods struggle to effectively utilize complex genetic and physiological network information.

Purpose of the Study:

• Introduce a theoretical framework for Artificial Intelligence (AI)-enabled prediction in crop improvement.
• Demonstrate the framework's capabilities and limitations using a logistic map model.

Main Methods:

• Integrated dynamical systems modeling, ensemble methods, Bayesian statistics, and optimization.
• Utilized a logistic map to simulate system complexity and assess predictability.
• Compared prediction of system states versus system process rates.

Main Results:

• Predictive skill increases with system complexity when using symbolic/sub-symbolic AI.
• Heritability and predictability decrease with increasing system complexity, aligning with empirical data.
• Predicting system process rates is more effective than predicting system states for complex systems.

Conclusions:

• AI-enabled prediction frameworks can overcome the curse of dimensionality in genomic prediction.
• Leveraging prior biological knowledge and computational approaches enhances prediction accuracy.
• This integrated approach promises to accelerate genetic gain in crop breeding programs.