CLCNet: a contrastive learning and chromosome-aware network for genomic prediction in plants

Area of Science:

• Plant breeding and genetics
• Bioinformatics and computational biology
• Machine learning in agriculture

Background:

• Genomic selection (GS) uses genome-wide markers to predict breeding values, but current models face challenges like capturing inter-individual variability and the curse of dimensionality.
• Existing genomic prediction (GP) methods struggle with high-dimensional SNP data and accurately identifying subtle genotype-phenotype relationships.

Purpose of the Study:

• To introduce CLCNet (Contrastive Learning and Chromosome-aware Network), a novel deep learning framework designed to enhance genomic prediction accuracy.
• To address limitations in current GP models by integrating contrastive learning and chromosome-aware feature selection.

Main Methods:

• CLCNet employs a contrastive learning module to discern fine-grained, genotype-dependent phenotypic differences among individuals.
• A chromosome-aware module facilitates structured feature selection at chromosome and genome levels, distilling the most informative single-nucleotide polymorphisms (SNPs).
• The framework was evaluated across four crop species and ten traits, compared against various classical and machine learning models.

Main Results:

• CLCNet demonstrated superior prediction performance, with statistically significant improvements in Pearson correlation coefficient (0.34%–12.19%) and reduced mean squared error compared to baseline models.
• Performance gains were most notable for traits with moderate linkage disequilibrium (LD) and high heritability in crops like maize, rapeseed, and soybean.
• The model maintained robust performance for cotton traits, even those with high LD and lower heritability.

Conclusions:

• CLCNet is an effective deep learning framework for improving genomic prediction accuracy in plant breeding.
• The integration of contrastive learning and chromosome-aware feature selection offers a powerful approach to overcome existing GP challenges.
• CLCNet shows significant potential for practical applications in accelerating crop improvement programs.