Deep Learning of Plant CIS Regulatory Code to Predict Expression and Trait Associated Variants

Abstract

Cis-regulatory elements (CREs) play a central role in controlling gene expression and shaping phenotypic diversity in plants, yet the regulatory code embedded within noncoding genomic regions remains incompletely understood. Recent advances in deep learning have enabled the direct prediction of transcriptional activity from DNA sequence, primarily in animal systems, with comparatively limited application to plant regulatory genomics. This work surveys and frames the use of deep learning approaches to decode the plant cis-regulatory code, predict gene expression, and prioritize trait-associated regulatory variants. We review the biological foundations of plant regulatory genomics, including promoters, enhancers, transcription factor networks, and noncoding variation, and discuss how cis-regulatory variants contribute disproportionately to phenotypic and agronomic traits. We further examine deep learning architectures-such as convolutional neural networks, transformers, and hybrid models- and their suitability for modeling sequence–function relationships in plant genomes. Emphasis is placed on feature representation, training paradigms, cross-species transfer, and evaluation strategies, as well as challenges arising from data heterogeneity, limited expression datasets, and environmental context dependence. By integrating curated genomic, transcriptomic, and trait datasets, deep learning–based regulatory models offer a promising path toward genome-to-phenome prediction, improved causal variant identification, and trait-informed crop improvement.