Enhancing genomic prediction with Stacking Ensemble Learning in Arabica Coffee

Area of Science:

• Plant Breeding and Genetics
• Bioinformatics and Computational Biology
• Agricultural Science

Background:

• Traditional coffee breeding relies on lengthy phenotypic observations, hindering rapid cultivar development.
• Genomic selection (GS) offers a faster, DNA-based alternative for identifying superior coffee (Coffea Arabica) genotypes.
• Ensemble learning methods, particularly Stacking Ensemble Learning (SEL), show potential for enhancing prediction accuracy in complex trait selection.

Purpose of the Study:

• To investigate the efficacy of Stacking Ensemble Learning (SEL) for improving prediction accuracy in Coffea Arabica genomic selection.
• To evaluate SEL's performance in predicting key agronomic and disease resistance traits: yield (YL), fruit number (NF), leaf miner infestation (LM), and cercosporiosis incidence (Cer).

Main Methods:

• Analysis of 195 Coffea Arabica individuals genotyped with 21,211 single-nucleotide polymorphism (SNP) markers.
• Implementation of a cross-validation (CV) scheme to assess model performance.
• Utilized Genomic Best Linear Unbiased Prediction (GBLUP), MARS, QRF, and RF as base learners within the SEL framework, with Ridge Regression, RF, GBLUP, and Single Average as meta-learners.

Main Results:

• Stacking Ensemble Learning (SEL) demonstrated superior predictive ability (PA) across all evaluated traits compared to individual base learner models.
• SEL achieved significant gains in PA over GBLUP: 87.44% for yield (YL), 37.83% for fruit number (NF), 199.82% for leaf miner infestation (LM), and 14.59% for cercosporiosis incidence (Cer).
• The study confirmed SEL's capability to accurately predict important traits in Coffea Arabica.

Conclusions:

• Stacking Ensemble Learning (SEL) represents a highly promising advancement for genomic selection in coffee breeding.
• By integrating predictions from multiple models, SEL effectively enhances the predictive accuracy for complex traits in Coffea Arabica.
• This approach accelerates the selection of superior coffee cultivars, addressing limitations of traditional breeding methods.