MSFF-CDCGAN: A novel method to predict RNA secondary structure based on Generative Adversarial Network.

Area of Science:

• Computational Biology
• Bioinformatics
• Machine Learning

Background:

• RNA secondary structures are crucial for cellular function but challenging to determine experimentally.
• Existing computational methods struggle with long RNA sequences and pseudoknots, representing a significant bottleneck.
• Deep learning approaches show promise but are hindered by insufficient data for complex RNA structures.

Purpose of the Study:

• To develop an accurate computational method for predicting RNA secondary structures, particularly long sequences and those with pseudoknots.
• To address the data scarcity issue in deep learning for RNA structure prediction.
• To introduce a novel deep learning architecture for RNA secondary structure analysis.

Main Methods:

• RNA data encoded into grayscale images using a unique encoding method.
• Image data augmentation techniques employed to expand the dataset.
• Development and application of a multi-scale feature fusion Conditional Deep Convolutional Generative Adversarial Network (MSFF-CDCGAN) model.

Main Results:

• The MSFF-CDCGAN model demonstrated higher accuracy in predicting long-sequence RNAs and pseudoknots compared to traditional methods.
• Successfully transformed RNA secondary structure prediction into an image analysis problem.
• Provided a robust data foundation for deep learning-based RNA structure prediction.

Conclusions:

• The proposed MSFF-CDCGAN model effectively overcomes the limitations of existing methods for RNA secondary structure prediction.
• This work pioneers the application of Generative Adversarial Networks (GANs) in RNA secondary structure prediction.
• The image-based data expansion and novel deep learning approach offer a promising solution for complex RNA structure analysis.