S2Snet: deep learning for low molecular weight RNA identification with nanopore

Area of Science:

• Bioinformatics
• Molecular Biology
• Computational Biology

Background:

• Ribonucleic acid (RNA) is crucial for biological regulation.
• Existing machine learning methods for RNA nanopore data require manual feature extraction, which is inefficient.
• Nanopore-generated RNA structural events often have unequal lengths, posing challenges for deep learning models.

Purpose of the Study:

• To develop a deep learning framework for automatic classification of RNA structural events from nanopore data.
• To address the challenge of unequal length sequences (UELS) in nanopore data.
• To improve the accuracy and efficiency of RNA structural event classification.

Main Methods:

• A sequence-to-sequence (S2S) module was developed to transform UELS into equal-length sequences.
• A deep learning-based sequence-to-sequence neural network was employed for automatic feature extraction.
• An attention mechanism was incorporated to capture critical features like dwell time and blockage amplitude.

Main Results:

• The proposed deep learning method achieved approximately a 2% increase in accuracy compared to previous machine learning approaches.
• The S2S module effectively handles UELS, making nanopore data compatible with deep learning models.
• The attention mechanism successfully identified key features for improved classification.

Conclusions:

• The novel deep learning approach offers a more efficient and accurate method for classifying RNA structural events from nanopore data.
• The proposed framework provides a generalizable strategy for processing nanopore data with UELS.
• This method has potential applications on various nanopore platforms, including Oxford Nanopore.