RPITER

RPITER predicts interactions between non-coding RNAs (ncRNAs) and proteins using a hierarchical deep learning framework to improve identification of RNA–protein interactions and support study of ncRNA function, including long non-coding RNAs (lncRNAs) and RNA-binding proteins.

Key Features:

• Hierarchical deep learning framework: Employs a hierarchical architecture combining convolutional neural networks (CNN) and stacked auto-encoders (SAE) to model sequence-derived features.
• Improved conjoint triad feature (CTF) coding: Uses an enhanced CTF encoding that incorporates additional primary sequence information and integrates sequence structural data.
• k-mer feature representation: Derives k-mer features from RNA and protein sequences for input to the deep models.
• CNN and SAE fitting capability: Both CNN and SAE architectures demonstrate robust capability in fitting k-mer features from sequences.
• Benchmark evaluation: Evaluated on five benchmark datasets sourced from the Protein Data Bank (PDB) and NPInter: RPI369, RPI488, RPI1807, RPI2241, and NPInter.
• Predictive performance: Reported area under the curve (AUC) scores of 0.821 (RPI369), 0.911 (RPI488), 0.990 (RPI1807), 0.957 (RPI2241), and 0.985 (NPInter).
• Comparative performance: Demonstrated performance that outperforms many existing methods across the evaluated datasets.
• RPI network construction: Enables construction of RNA–protein interaction (RPI) networks for downstream analysis of ncRNA and lncRNA function.

Scientific Applications:

• Predicting ncRNA–protein interactions: Prioritizes candidate interactions between ncRNAs (including lncRNAs) and RNA-binding proteins.
• RPI network analysis: Supports building RNA–protein interaction networks for functional and systems-level analysis.
• Study of post-transcriptional regulation and disease: Facilitates investigation of ncRNA roles in post-transcriptional gene regulation and human disease mechanisms.

Methodology:

Uses an improved conjoint triad feature (CTF) coding that incorporates primary sequence and sequence structural information, extracts k-mer features from RNA and protein sequences, applies a hierarchical deep learning framework combining convolutional neural networks (CNN) and stacked auto-encoders (SAE), and evaluates performance on RPI369, RPI488, RPI1807, RPI2241 and NPInter benchmark datasets.