Construction of gene regulatory networks (GRNs) is essential for elucidating the regulatory mechanisms underlying metabolic pathways, biological processes, and complex traits. In this study, we developed and evaluated machine learning, deep learning, and hybrid approaches for constructing GRNs by integrating prior knowledge and large-scale transcriptomic data from Arabidopsis thaliana, poplar, and maize. Among these, hybrid models that combined convolutional neural networks and machine learning consistently outperformed traditional machine learning and statistical methods, achieving over 95% accuracy on the holdout test datasets. These models not only identified a greater number of known transcription factors regulating the lignin biosynthesis pathway but also demonstrated higher precision in ranking key master regulators such as MYB46 and MYB83, as well as many upstream regulators, including members of the VND, NST, and SND families, at the top of candidate lists. To address the challenge of limited training data in non-model species, we implemented transfer learning, enabling cross-species GRN inference by applying models trained on well-characterized and data-rich species to another species with limited data. This strategy enhanced model performance and demonstrated the feasibility of knowledge transfer across species. Overall, our findings underscore the effectiveness of hybrid and transfer learning approaches in GRN prediction, offering a scalable framework for elucidating regulatory mechanisms in both model and non-model plant systems.
The construction of gene regulatory networks (GRNs) is vital for understanding the regulation of metabolic pathways, biological processes, and complex traits during plant growth and responses to environmental cues and stresses. The increasing availability of public databases has facilitated the development of numerous methods for inferring gene regulatory relationships between transcription factors and their targets. However, there is limited research on supervised learning techniques that utilize available regulatory relationships of plant species in public databases. This study investigates the potential of machine learning (ML), deep learning (DL), and hybrid approaches for constructing GRNs in plant species, specifically Arabidopsis thaliana, poplar, and maize. Challenges arise due to limited training data for gene regulatory pairs, especially in less-studied species such as poplar and maize. Nonetheless, our results demonstrate that hybrid models integrating ML and artificial neural network (ANN) techniques significantly outperformed traditional methods in predicting gene regulatory relationships. The best-performing hybrid models achieved over 95% accuracy on holdout test datasets, surpassing traditional ML and ANN models and also showed good accuracy on lignin biosynthesis pathway analysis. Employing transfer learning techniques, this study has also successfully transferred the known knowledge of gene regulation from one species to another, substantially improving performance and manifesting the viability of cross-species learning using deep learning-based approaches. This study contributes to the methodology for growing body of knowledge in GRN prediction and construction for plant species, highlighting the value of adopting hybrid models and transfer learning techniques. This study and the results will help to pave a way for future research on how to learn from known to unknown and will be conductive to the advance of modern genomics and bioinformatics.