RGPA-GCN: Graph convolutional networks for rice gene-phenotype association prediction.

Area of Science:

• Agricultural Science
• Bioinformatics
• Computational Biology

Background:

• Rice is a crucial global food source, and understanding gene-phenotype associations (GPAs) is key to enhancing its stress tolerance and yield.
• Experimental identification of GPAs is resource-intensive and time-consuming.
• Computational methods offer an efficient alternative to accelerate GPA discovery.

Purpose of the Study:

• To develop an innovative computational approach for predicting gene-phenotype associations (GPAs) in rice.
• To leverage graph convolutional networks (GCNs) for accurate GPA prediction.
• To create a tool that can identify both known and novel GPAs, including those involving previously uncharacterized genes or phenotypes.

Main Methods:

• Framed GPA prediction as a node classification task.
• Introduced RGPA-GCN, a graph convolutional network model.
• Constructed a topology graph using the k-nearest neighbor method for information aggregation.
• Incorporated gene functional similarity and phenotype semantic similarity into graph nodes.

Main Results:

• RGPA-GCN demonstrated strong performance in predicting GPAs.
• The model outperformed six classical machine learning methods and three state-of-the-art models in 5-fold cross-validation.
• Ablation studies and case studies on five phenotypes confirmed the model's effectiveness.
• The approach successfully predicted unknown GPAs and novel genes/phenotypes.

Conclusions:

• RGPA-GCN is an effective computational tool for predicting gene-phenotype associations in rice.
• This method significantly accelerates the discovery of GPAs, aiding in the development of improved rice varieties.
• The model's ability to predict novel associations highlights its potential for future genetic research in rice.