DeepBtoD: Improved RNA-binding proteins prediction via integrated deep learning.

Area of Science:

• Computational Biology
• Bioinformatics
• Molecular Biology

Background:

• RNA-binding proteins (RBPs) are vital for cellular processes like gene regulation and alternative splicing.
• Accurate identification of RBPs is crucial for understanding these biological functions.
• Existing computational methods often lack simultaneous consideration of local and global RNA sequence information.

Purpose of the Study:

• To develop a novel computational method, DeepBtoD, for predicting RBPs directly from RNA sequences.
• To integrate both local and global sequence features for improved RBP prediction accuracy.
• To provide a valuable tool for modeling protein-nucleic acid interactions in systems biology.

Main Methods:

• A novel [Formula: see text]-BtoD encoding method was designed to capture local nucleotide composition and relative positions.
• A multi-scale convolutional neural network (ms-focusCNN) with a self-attentive mechanism was employed to learn high-level sequence features.
• Ensemble learning combined local and global information for the final RBP prediction.

Main Results:

• DeepBtoD achieved an area under the curve (AUC) of 0.933 on 24 independent test datasets.
• The method demonstrated competitive performance against seven state-of-the-art RBP prediction tools.
• Integration of local [Formula: see text]-BtoD encoding and global information significantly improved RBP recognition.

Conclusions:

• DeepBtoD offers a powerful and accurate approach for predicting RBPs solely from RNA sequences.
• The method's ability to integrate diverse sequence information enhances its predictive capabilities.
• DeepBtoD has the potential to advance systems biology research by improving the modeling of protein-nucleic acid interactions.