Predicting adult phenotypes from seedling transcriptional data using deep learning: a case study in chrysanthemum

Abstract

Genotype-to-phenotype prediction remains a fundamental challenge in current genetic research. In recent years, it has become possible to construct different predictive models based on genomic data. However, in many horticultural crops, it is difficult to accurately verify genomic variations because of the complexity of their genome, making the application of these genome-based methods challenging. Gene expression reflects both genetic regulatory mechanisms and environmental stimuli, offering potential for predicting phenotypes in plants with complex genomes. Thus, in this paper, we tested the possibility for predicting adult plant phenotypes using the gene expression data from seedlings. By applying the transcriptional-based deep learning methods on cut chrysanthemums (Chrysanthemum spp.), which exhibits a complex genetic background characterized by high repetitiveness, heterozygosity, and genome size and is recognized as a segmental allopolyploid, we found that the method is robust and accurate for predicting continuous variables such as leaf vase life, as well as categorical variables such as flower types on the basis of gene expression data. Moreover, the power and performance of transcriptional-based deep learning methods for prediction was validated in rice (Oryza sativa). Our research shows the good performance of phenotype prediction based on gene expression, with potential applications in future gene chip-based breeding practices.

Plant phenotyping relevance

遺伝子発現データから成体の植物形質を予測する深層学習手法を開発・検証しており、形質予測が研究の中心である。

Code and data availability

The paper deposits its authors' analysis code publicly on GitHub and its raw RNA-seq data (used for the seedling-transcriptome phenotype prediction) in the Genome Sequence Archive with accession CRA022074. Both are paper-specific, public, and actionable.