Research on Plant RNA-Binding Protein Prediction Method Based on Improved Ensemble Learning

Area of Science:

• Plant molecular biology
• Bioinformatics
• Computational biology

Background:

• RNA-binding proteins (RBPs) are essential regulators of gene expression in plants, influencing growth, development, and stress adaptation.
• Accurate identification of plant-specific RBPs is critical for advancing our understanding of gene regulation and for genetic improvement strategies in agriculture.

Purpose of the Study:

• To develop and validate a highly accurate computational method for predicting plant RNA-binding proteins.
• To enhance the prediction of plant RBPs by integrating shallow and deep learning approaches.

Main Methods:

• An ensemble learning framework was developed, combining Support Vector Machine (SVM), Logistic Regression (LR), Linear Discriminant Analysis (LDA), and LightGBM predictions with an enhanced TextCNN model.
• K-Peptide Composition (KPC) encoding with k=1, 2 generated a 420-dimensional feature vector, extended to 424 dimensions with ensemble outputs.
• Feature redundancy was minimized using a Pearson correlation threshold of 0.80.

Main Results:

• The proposed method achieved high accuracy on a benchmark dataset (97.20% ACC with 5-fold cross-validation) and an independent dataset (99.72% ACC, 99.72% F1-score, 99.45% MCC).
• Performance significantly surpassed existing methods, outperforming RBPLight by 12.98% and the original TextCNN by 25.23% in accuracy.
• The method demonstrated superior efficiency and accuracy compared to PSSM-based approaches.

Conclusions:

• The developed ensemble learning method offers a powerful and accurate tool for large-scale prediction of plant RNA-binding proteins.
• This advancement facilitates deeper insights into plant gene regulation and supports efforts in crop genetic improvement.