Editorial: Machine Learning for Mining Plant Functional Genes

Shanwen Sun¹Quan Zou^2*Lijun Dou^3*

• ¹Northeast Forestry University, Harbin, China
• ²University of Electronic Science and Technology of China, Chengdu, China
• ³Cleveland Clinic, Cleveland, United States

representations that capture latent regulatory and functional features. Compared with task-specific models, FMs offer enhanced generalization, cross-species transferability, and scalability, making them particularly attractive for plant systems characterized by genomic complexity and limited functional annotations. Providing a systematic overview of this paradigm, Xu et al. present a mini-review synthesizing recent advances in foundation models for plant molecular biology. The review traces the evolution from general DNA language models to plant-specific tools and highlights key challenges unique to plant systems, including polyploidy, repetitive genomes, and sparse experimental annotations. By outlining future directions such as multimodal integration and computational efficiency, this work establishes a conceptual framework for understanding how FMs are redefining computational plant biology and guiding next-generation model development.Building on this FM paradigm, several contributions demonstrate how representation learning can be applied to concrete biological problems. Zhang et al. applied a DNABERT-2-based framework combined with gradient boosting to identify DNA N6-methyladenine modifications in rice, illustrating how foundation models can enhance epigenetic marker detection while mitigating data sparsity. This work exemplifies a broader shift toward pretraining-based strategies in plant genomics, with implications for cross-species prediction and regulatory annotation. Taken together, the contributions in this Research Topic highlight the transformative role of machine learning and foundation models in plant functional genomics. By advancing representation learning, model architecture, interpretability, and multi-omics integration, these studies move the field beyond traditional sequence-based annotation toward predictive, mechanism-aware, and application-oriented frameworks. Continued synergy between computational innovation and experimental validation will be essential for translating these advances into resilient, high-yield crops capable of meeting future agricultural and environmental challenges.