Predicting rice drought-responsive genes via distance-based prototypical graph neural network with path aggregation...

Area of Science:

• Genomics
• Bioinformatics
• Plant Science

Background:

• Drought stress significantly reduces rice yield and quality.
• Identifying drought-responsive genes is critical for developing resilient rice cultivars.
• Existing graph neural network methods face challenges in capturing gene attributes, network topology, and class imbalance.

Purpose of the Study:

• To propose a novel distance-based prototypical graph neural network with path aggregation mechanism (DPGNNPAM) for identifying drought-responsive genes in rice.
• To enhance the predictive capability of graph neural networks in biological networks by addressing limitations in feature representation and class imbalance.

Main Methods:

• Constructed graph-based datasets by integrating rice gene expression data and protein interaction networks.
• Employed a random walk strategy and a recursive neural network-based path aggregator to encode node attributes along diverse paths.
• Utilized a prototypical network approach to focus on global information, address sample imbalance, and compute weighted similarity between node embeddings and class prototypes.

Main Results:

• The proposed DPGNNPAM model demonstrated superior performance compared to traditional graph neural network algorithms in identifying drought-responsive genes.
• Identified 17 candidate genes associated with drought stress in rice.
• Validated 12 of the identified candidate genes through existing scientific literature, confirming their role in drought stress response.

Conclusions:

• DPGNNPAM effectively mines drought-responsive genes in rice by integrating network topology and node attributes while managing class imbalance.
• The identified genes provide valuable targets for breeding drought-tolerant rice varieties.
• This study highlights the potential of advanced graph neural network methods in plant genomics and stress response research.