Enhancing Taxonomic Categorization of DNA Sequences with Deep Learning: A Multi-Label Approach.

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning

Background:

• Accurate taxonomic categorization of DNA sequences is crucial for biological research.
• Traditional methods may struggle with the complexity and scale of genomic data.
• Deep learning offers potential for enhanced sequence analysis and classification.

Purpose of the Study:

• To investigate the application of deep learning for DNA sequence taxonomic categorization.
• To propose and evaluate two novel deep learning architectures: Stacked Convolutional Autoencoder (SCAE)-MLELM and Variational Convolutional Autoencoder (VCAE)-MLELM.
• To assess the impact of incorporating multiple taxonomic labels on classification accuracy.

Main Methods:

• Development of SCAE-MLELM and VCAE-MLELM architectures for feature extraction and classification.
• Utilizing Multilabel Extreme Learning Machine (MLELM) for processing extracted features and generating classification scores.
• Training and testing models on unsupervised DNA sequence data, considering single and multiple labels concurrently.

Main Results:

• The VCAE-MLELM model consistently outperformed the SCAE-MLELM model across all tested conditions.
• Incorporating the clade label significantly improved accuracy for both models compared to class or genus labels.
• The highest accuracy achieved was 94% using the VCAE-MLELM model with combined clade and family labels.
• Single-label categorization accuracy for both models remained below 65%.

Conclusions:

• Deep learning models, particularly VCAE-MLELM, show strong potential for accurate DNA sequence taxonomic classification.
• Combining multiple taxonomic labels (e.g., clade, family) is essential for maximizing classification performance.
• The MLELM network's ability to capture inter-class patterns is key to the approach's effectiveness in biological taxonomy.