EDCNN: identification of genome-wide RNA-binding proteins using evolutionary deep convolutional neural network

Area of Science:

• Computational biology
• Bioinformatics
• Genomics

Background:

• RNA-binding proteins (RBPs) are crucial for RNA regulation, metabolism, and processing.
• Existing computational methods for predicting RBPs face challenges like high dimensionality, data sparsity, and suboptimal performance.

Purpose of the Study:

• To enhance the performance of deep convolutional neural networks for predicting RNA-binding events.
• To introduce a novel computational method, Evolutionary Deep Convolutional Neural Network (EDCNN), for identifying protein-RNA interactions.

Main Methods:

• EDCNN synergizes evolutionary optimization with gradient descent for enhanced deep learning.
• The algorithm employs alternating gradient descent and evolution steps to optimize RNA-binding event detection.
• Validated on large-scale CLIP-seq datasets.

Main Results:

• EDCNN demonstrates superior performance compared to existing state-of-the-art methods.
• Experimental results confirm the effectiveness of EDCNN in identifying protein-RNA interactions.
• Further analyses (time complexity, parameter, motif) support the algorithm's robustness.

Conclusions:

• EDCNN offers a significant advancement in predicting RNA-binding proteins and their interactions.
• The proposed method effectively addresses limitations of previous computational approaches.
• EDCNN provides a powerful tool for RNA regulation and metabolism research.