Crop phenotype prediction using biclustering to explain genotype-by-environment interactions.

Area of Science:

• Plant genetics
• Agricultural science
• Statistical modeling

Background:

• Phenotypic variation in plants arises from genotype (G), environment (E), and their interaction (GEI).
• GEI is crucial for understanding consistent genotype performance across diverse environments.
• Plant breeding often faces incomplete data, with genotypes tested in limited environments.

Purpose of the Study:

• To propose a novel model for genotype-by-environment interaction (GEI) effects that accommodates missing data.
• To develop a method for identifying subsets of genotypes and environments with simplified phenotype interactions.
• To predict crop performance in untested environments using a novel biclustering approach.

Main Methods:

• Utilized a novel biclustering algorithm capable of handling missing values in phenotype observation tables.
• Developed a GEI model requiring only a two-way table of phenotype observations (genotypes vs. environments).
• Fitted no-interaction models to predict phenotypes and assess cultivar performance in unobserved environments.

Main Results:

• The proposed biclustering algorithm successfully identifies homogeneous cells with no GEI.
• The model effectively handles incomplete two-way tables of phenotype observations.
• Validated methodology shows superior performance over existing statistical approaches across various plant species and phenotypes.

Conclusions:

• The new GEI model provides a robust solution for analyzing incomplete plant breeding data.
• Accurate prediction of cultivar performance in new environments is achievable.
• This approach offers significant advantages for plant breeding and crop improvement strategies.