RPI-EDLCN: An Ensemble Deep Learning Framework Based on Capsule Network for ncRNA-Protein Interaction Prediction

Area of Science:

• Genomics and Molecular Biology
• Computational Biology
• Bioinformatics

Background:

• Noncoding RNAs (ncRNAs) are critical regulators of cellular processes through interactions with proteins.
• Identifying ncRNA-protein interactions (ncRPIs) is essential for elucidating ncRNA functions.
• Existing computational methods for ncRPI prediction face challenges in feature extraction and model architecture.

Purpose of the Study:

• To develop an advanced computational framework for accurate prediction of ncRNA-protein interactions (ncRPIs).
• To improve the recognition performance of ncRPI prediction models by integrating diverse feature types and a novel deep learning architecture.

Main Methods:

• Proposed RPI-EDLCN, an ensemble deep learning framework utilizing Capsule Network (CapsuleNet).
• Extracted sequence, secondary structure, motif, and physicochemical properties of ncRNA/protein.
• Encoded features using conjoint k-mer and processed them via CNN, DNN, and SAE before CapsuleNet learning.

Main Results:

• RPI-EDLCN achieved superior performance compared to state-of-the-art methods across multiple datasets (e.g., 93.8% accuracy on RPI1807).
• Demonstrated effective prediction of potential ncRPIs in different organisms via independent testing.
• Successfully identified hub ncRNAs and proteins in Mus musculus ncRNA-protein networks.

Conclusions:

• RPI-EDLCN serves as a robust and effective tool for predicting ncRNA-protein interactions.
• The framework provides valuable insights for future biological studies on ncRNA functions and networks.
• Highlights the potential of ensemble deep learning with CapsuleNet for complex biological interaction predictions.