A survey on deep learning in DNA/RNA motif mining

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• DNA/RNA motif mining is crucial for understanding gene regulation by identifying DNA- or RNA-protein binding sites.
• Traditional motif mining algorithms include enumeration and probabilistic methods.
• Machine learning, particularly deep learning, has recently shown significant advancements in motif mining accuracy and efficiency.

Purpose of the Study:

• To review and summarize the application of deep learning in DNA/RNA motif mining.
• To analyze data preprocessing, deep learning architectures, and compare different models.
• To highlight the potential of deep learning for improving motif discovery.

Main Methods:

• Categorization of existing deep learning methods into convolutional neural network (CNN), recurrent neural network (RNN), and hybrid CNN-RNN models.
• Analysis of data preprocessing techniques relevant to motif mining.
• Comparative study of different deep learning architectures and their performance characteristics.

Main Results:

• Deep learning models, especially complex ones, demonstrate superior performance in motif mining when ample data is available.
• Current deep learning approaches in motif mining are less complex compared to those in fields like computer vision or NLP.
• Convolutional neural network (CNN) and recurrent neural network (RNN) based models are key architectures in this domain.

Conclusions:

• Deep learning offers a powerful approach to enhance DNA/RNA motif mining for gene function research.
• Further development and increased model complexity in deep learning are warranted for motif mining.
• This review provides a foundational understanding for researchers entering the field of deep learning-based motif mining.