DNASimCLR: a contrastive learning-based deep learning approach for gene sequence data classification.

Area of Science:

• Genomics
• Bioinformatics
• Machine Learning

Background:

• Deep neural networks advance microbial sequence data analysis.
• Labeled microbial data scarcity hinders supervised learning.
• Unsupervised learning is crucial for complex biological data.

Purpose of the Study:

• Introduce DNASimCLR, an unsupervised framework for gene sequence feature extraction.
• Address limitations of supervised learning with limited labeled microbial data.
• Enhance the analysis of microbial sequence data.

Main Methods:

• Utilize convolutional neural networks (CNNs) and the SimCLR framework.
• Employ contrastive learning for feature extraction from diverse microbial gene sequences.
• Pre-train on large-scale unlabeled metagenome and viral gene datasets.

Main Results:

• DNASimCLR achieves performance comparable to state-of-the-art methods.
• Outperforms existing CNN-based feature extraction techniques.
• Demonstrates robust adaptability and superior performance across various biological sequence analysis tasks.

Conclusions:

• DNASimCLR provides a robust, database-agnostic solution for gene sequence classification.
• Effective for novel and unseen gene sequences, valuable for diverse genomics applications.
• Advances unsupervised learning in microbial genomics.