ExAutoGP: Enhancing Genomic Prediction Stability and Interpretability with Automated Machine Learning and SHAP

Area of Science:

• Genomics
• Machine Learning
• Bioinformatics

Background:

• Genomic prediction models are powerful but often lack transparency.
• Interpreting machine learning decisions is crucial for biological insights.

Purpose of the Study:

• To introduce ExAutoGP, a novel method combining AutoML and SHAP for improved genomic prediction.
• To enhance model interpretability and identify key genetic markers.

Main Methods:

• ExAutoGP was developed by integrating Automated Machine Learning (AutoML) with SHapley Additive exPlanations (SHAP).
• Comparative experiments used simulated and real animal datasets, evaluating ExAutoGP against GBLUP, BayesB, SVR, KRR, and RF.
• Performance was assessed using five repeated five-fold cross-validation, focusing on predictive accuracy and computational efficiency.

Main Results:

• ExAutoGP demonstrated robust and superior prediction performance across all tested datasets.
• SHAP analysis effectively elucidated ExAutoGP's decision-making process, enhancing interpretability.
• Key genetic markers associated with specific traits were successfully identified.

Conclusions:

• AutoML shows significant potential for advancing genomic prediction.
• The integration of SHAP provides valuable, actionable biological insights.
• The combination of accuracy and interpretability offers new strategies for genomic selection in breeding programs.