Protein-Binding RNA Prediction Based on Integrated Sequence-Structure-Function Pre-Training.

Area of Science:

• Computational biology
• Molecular biology
• Bioinformatics

Background:

• RNA binding proteins (RBPs) are essential regulators of biological functions.
• High-throughput experiments generate vast RNA interaction data.
• Existing computational methods face challenges in integrating multi-level semantic information and predicting accurately on small datasets.

Purpose of the Study:

• To develop an advanced computational method, MTP-RBP, for improved RNA binding protein prediction.
• To enhance the integration of multi-level semantic information from RNA sequences.
• To boost predictive accuracy, particularly for small-sample datasets.

Main Methods:

• MTP-RBP employs multi-task pre-training with a robust pre-trained encoder.
• It extracts deep contextual information from RNA sequences.
• Incorporates structural and functional RNA knowledge through masked language modeling, secondary structure construction, and binding function prediction pre-training tasks.

Main Results:

• MTP-RBP demonstrates state-of-the-art performance in RBP prediction.
• It surpasses existing baseline and RNA language models.
• Achieves superior accuracy, especially on small datasets.

Conclusions:

• MTP-RBP effectively fuses multi-level RNA features for enhanced prediction.
• The method offers a significant advancement in computational RBP identification.
• It provides a more accurate and robust tool for analyzing RNA-protein interactions.