RPITER: A Hierarchical Deep Learning Framework for ncRNA⁻Protein Interaction Prediction.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Non-coding RNAs (ncRNAs) are critical regulators of gene expression and are implicated in human diseases.
• Understanding ncRNA-protein interactions (RPIs) is essential for elucidating ncRNA functions.
• Experimental RPI identification methods are costly and time-consuming, necessitating computational approaches.

Purpose of the Study:

• To develop an accurate and efficient computational method for predicting RNA-protein interactions (RPIs).
• To improve sequence coding strategies for RNA and protein sequences in RPI prediction.
• To leverage deep learning architectures for enhanced RPI prediction performance.

Main Methods:

• Proposed RPITER, a hierarchical deep learning framework for RPI prediction.
• Enhanced the conjoint triad feature (CTF) coding method with primary sequence and structure information.
• Utilized convolution neural network (CNN) and stacked auto-encoder (SAE) architectures for feature learning.

Main Results:

• RPITER demonstrated strong performance in predicting RPIs across five benchmark datasets (PDB, NPInter).
• The improved CTF coding method and deep learning architectures (CNN, SAE) significantly enhanced prediction accuracy.
• RPITER achieved high AUC values, outperforming most existing RPI prediction methods.

Conclusions:

• RPITER provides a robust and effective computational tool for predicting RPIs.
• The developed method can complement experimental techniques for RPI identification and network construction.
• This work facilitates further research into ncRNA and long non-coding RNA (lncRNA) functions and disease mechanisms.