Root-associated protein prediction using a protein large language model and hypergraph convolutional networks.

Area of Science:

• Plant Biology
• Bioinformatics
• Computational Biology

Background:

• Plant root-associated proteins are crucial for growth, signaling, and stress tolerance.
• Identifying these proteins is vital for plant science but current methods are costly and slow.
• Existing computational models for protein prediction require significant improvement.

Purpose of the Study:

• To develop an advanced computational model for predicting plant root-associated proteins.
• To overcome limitations of traditional experimental and current computational approaches.
• To enhance the discovery of novel root-associated proteins.

Main Methods:

• Developed a novel computational model named Hypergraph-Root.
• Utilized protein features from BLOSUM62, position-specific scoring matrices, and a protein language model.
• Enhanced feature representation using hypergraph convolutional networks and multi-head attention.
• Employed a fully connected layer for final prediction.

Main Results:

• The Hypergraph-Root model achieved high prediction performance with an AUC of approximately 0.9 on training and independent datasets.
• Demonstrated significant advantages over existing computational models.
• Experimental validation confirmed the model's structural rationality.

Conclusions:

• Hypergraph-Root offers a powerful and efficient computational approach for predicting plant root-associated proteins.
• The model significantly advances the identification of proteins involved in plant growth and environmental adaptation.
• This work provides a valuable tool for future research in plant biology and crop improvement.