#32 · Genomics

Hyperspectral latent factors and deep learning extend plant genomic prediction

The dominant thread is prediction in plants that combines genotype with additional context: phenotype dynamics, environment, or cross-species sequence information. One preprint factors hyperspectral time-series phenotypes into latent variables using factor analysis with Procrustes rotation, which improves within-trial genomic prediction of wheat grain yield. A published study applies CNN and GBLUP models to winter wheat yield across 31 sites, modelling G×E to select environment-adapted hybrids. A separate evaluation of Sei-based deep learning reports cross-species histone mark prediction in plants, with family-level models generalizing to unprofiled Poaceae. What is new is the pairing of phenotypic latent structure and explicit environmental interaction modelling with regulatory-state prediction that transfers across species. These studies share a focus on predictive generalization rather than a single shared mechanism.