The gradient boosting algorithm and random boosting for genome-assisted evaluation in large data sets.

Area of Science:

• Animal Genetics
• Machine Learning
• Bioinformatics

Background:

• Genomic evaluation methods face challenges with increasing genetic variants and sample sizes.
• High-density single nucleotide polymorphism (SNP) arrays and genome sequencing generate large datasets.
• The boosting algorithm is a machine learning technique that can handle large datasets.

Purpose of the Study:

• To introduce and evaluate a modified boosting algorithm, termed 'random boosting', for genomic evaluation.
• To improve predictive ability and decrease computation time for genome-assisted evaluation in large datasets.
• To assess the performance of random boosting compared to the original boosting algorithm.

Main Methods:

• A modified boosting algorithm (random boosting) was developed using random marker selection.
• The algorithm was applied to a real dataset of 1,797 bulls with 39,714 SNPs for yield and type traits.
• A 2-fold cross-validation was implemented, comparing training and testing samples of sires based on birth year.

Main Results:

• Random boosting reduced computation time by 99% compared to the original boosting algorithm.
• Negligible differences in predictive ability (Pearson correlations) and bias were observed between the two algorithms.
• The modified algorithm demonstrated high accuracy and low bias in predicting phenotypes.

Conclusions:

• Random boosting is an efficient modification for accelerating genome-assisted evaluation in large-scale genomic datasets.
• This approach is suitable for handling data from large consortiums and complex genetic evaluations.
• The method offers a practical solution for faster genomic predictions without compromising accuracy.