Interspecies predictions of growth traits from quantitative transcriptome data acquired during fruit development.

Linking genotype and phenotype is a fundamental challenge in biology. In this respect, machine learning is playing a pivotal role in systems biology. As central phenotypic traits, fruit development and relative growth rate (RGR) result from interactions between gene regulation, metabolism, and environment. In the present study, we carried out a multispecies transcriptomic analysis of nine different fruits. To illustrate fruit transcriptomes, transcripts were first compared using multivariate methods, revealing similar main profiles. They were then used as variables to predict four growth traits, that is RGR, developmental progress, fruit weight, and protein content, using generalized linear models to decipher the mechanisms involving gene expression in development. The predictions were highly satisfactory despite disparities when the model did not include the entire panel of fruit species. Based on orthogroups derived from BLAST and annotated consensus sequences from gene ontology terminology, variables annotated for metabolic processes, especially those involving cell wall carbohydrates and proteins, were found to be the most effective in predicting growth. In addition, predictions were improved for RGR when introducing a 7 d lag between transcript contents and growth traits, suggesting the necessity of considering the proteins produced to enhance phenotypic trait predictions. These original results showed that growth traits can be predicted very well with generalized linear models based on orthogroups from multi-species transcriptomes.