Genome-Enabled Prediction Methods Based on Machine Learning.

Area of Science:

• Genomics
• Bioinformatics
• Machine Learning

Background:

• Artificial intelligence and machine learning (ML) have seen significant growth.
• ML algorithms enable computers to learn from data for prediction and classification.
• Genome-wide prediction (GWP) utilizes ML for analyzing genomic data.

Purpose of the Study:

• To describe semiparametric and nonparametric ML algorithms used in GWP.
• To evaluate the predictive performance of various ML algorithms in GWP.
• To identify the most robust and accurate ML methods for GWP.

Main Methods:

• A meta-analysis was conducted on 34 comparative ML studies from the last decade.
• A Thurstonian model was employed to evaluate algorithm predictive qualities.
• Focus on semiparametric and nonparametric ML algorithms for GWP in plants and animals.

Main Results:

• Kernel, Bayesian, and ensemble ML methods demonstrated superior robustness and predictive ability.
• These advanced methods showed higher accuracy in genome-wide based prediction tasks.
• Algorithm performance varied based on study type and data distribution.

Conclusions:

• Specific ML algorithms, including kernel, Bayesian, and ensemble methods, are highly effective for GWP.
• The choice of the optimal ML model depends on the specific study context and data characteristics.
• Further consideration of study design and data distribution is crucial for successful GWP.