DFpin: Deep learning-based protein-binding site prediction with feature-based non-redundancy from RNA level

Area of Science:

• Computational biology
• Molecular biology
• Bioinformatics

Background:

• Protein-RNA interactions are vital in human diseases.
• Identifying these interactions aids in computer-aided drug design.
• Current methods face challenges with sample similarity due to RNA binding site aggregation.

Purpose of the Study:

• To present DFpin, a novel method for predicting protein-interacting nucleotides in RNA.
• To address sample similarity issues in RNA sequence analysis.
• To improve the accuracy of identifying RNA regions involved in protein binding.

Main Methods:

• Utilized a redundancy method based on feature similarity and RNA mono-nucleotide composition to maintain sample diversity.
• Employed a deep forest model with a cascade structure for feature extraction and prediction.
• Implemented feature redundancy removal to avoid overfitting and retain key nucleotide sites.

Main Results:

• DFpin achieved 85.4% accuracy and a 93.3% area under the curve (AUC) in predicting protein-interacting nucleotides.
• The method demonstrated superior accuracy compared to existing approaches.
• Feature-based redundancy removal and the deep forest model proved effective for this prediction task.

Conclusions:

• DFpin offers an effective solution for predicting protein-interacting nucleotides in RNA.
• The combination of feature redundancy removal and deep forest models enhances prediction accuracy.
• This approach has significant implications for understanding disease mechanisms and advancing drug discovery.