AI Framework Links Genes, Environment to Leaf Vein Architecture

Leaf veins exhibit remarkable diversity in their architecture and patterning, yet previous gene-environment association studies often simplified their analysis by quantifying venation using only a small collection of low-dimensional summary traits, thereby discarding much of the rich structural information present in the original images. This new research introduces a comprehensive, integrated deep learning and statistical framework that represents the complete leaf vascular architecture as a whole-network image phenotype. The proposed framework employs a fine-tuned deep learning model, Edge Detection with Transformers (EDTER), to accurately extract the intricate whole-network leaf vascular architecture directly from RGB images. This is achieved by jointly learning local and global contextual features. To facilitate this, a new annotated leaf image database was constructed by integrating edge maps generated by DiffusionEdge with the Berkeley Segmentation Database (BSDS500). For the statistical component, the framework applies Semiparametric Sparse Canonical Correlation Analysis (SSCCA). This advanced method performs variable selection and models associations between repeatedly measured high-dimensional bivariate image responses and high-dimensional predictors, while simultaneously accommodating sparse, zero-inflated data represented by edge maps through a truncated latent Gaussian copula model. Applied to a real Populus dataset, the framework identified three significant gene-geography interactions associated with leaf vascular architecture, providing new biological insights and establishing a broadly applicable methodological framework for high-dimensional complex image phenotypes.