prPred-DRLF: Plant R protein predictor using deep representation learning features

Area of Science:

• Plant pathology
• Computational biology
• Bioinformatics

Background:

• Plant resistance (R) proteins are crucial for detecting pathogen invasion.
• Accurate prediction of R proteins is vital for plant disease management.
• Traditional R protein prediction methods rely on feature extraction, limiting performance.

Purpose of the Study:

• To develop a high-performance computational approach for plant R protein prediction.
• To leverage deep representation learning for improved amino acid encoding.
• To enhance the accuracy and efficiency of R protein identification.

Main Methods:

• Proposed prPred-DRLF, a novel computational approach for plant R protein prediction.
• Employed deep representation learning, specifically Bidirectional Long Short-Term Memory (BiLSTM) and Unified Representation (UniRep) embeddings.
• Utilized a Light Gradient Boosting Machine (LGBM) classifier for prediction.

Main Results:

• Fused BiLSTM and UniRep embeddings demonstrated superior performance over other features.
• The prPred-DRLF model achieved high accuracy (0.956), F1-score (0.933), and AUC (0.997) on an independent test set.
• Outperformed state-of-the-art methods (prPred, HMMER) in accuracy, F1-score, AUC, and recall.

Conclusions:

• prPred-DRLF offers a significant advancement in plant R protein prediction.
• The tool integrates deep representation learning for enhanced accuracy.
• A user-friendly webserver and downloadable script are available for biological research.