Learning, visualizing and exploring 16S rRNA structure using an attention-based deep neural network.

Area of Science:

• Computational Biology and Bioinformatics
• Microbial Genomics
• Machine Learning in Biology

Background:

• Recurrent neural networks with memory and attention are crucial for sequential data in natural language processing.
• Microbial DNA sequence analysis often requires complex pre-processing and manual interpretation for classification and phenotype prediction.
• Existing methods struggle to efficiently capture both short and long-term dependencies in microbial sequence data.

Purpose of the Study:

• To propose and evaluate an integrated deep learning model for microbial DNA sequence analysis.
• To predict taxonomic classifications and sample-associated attributes, including host phenotype, at the read/sequence level.
• To develop a novel attention-based deep network architecture, Read2Pheno, for enhanced microbiome sequence classification.

Main Methods:

• Developed an integrated deep learning model combining convolutional neural networks, recurrent neural networks, and attention mechanisms.
• Applied the model, Read2Pheno, to amplicon sequences, specifically 16S ribosomal RNA (rRNA) marker genes.
• Utilized attention mechanisms to identify informative nucleotide regions and encode sequences into meaningful vector representations.

Main Results:

• Read2Pheno achieved read-level phenotypic prediction on microbial DNA sequences.
• The attention layer automatically identified informative nucleotide regions, aiding classification.
• Aggregated read-level predictions robustly classified microbial communities and host phenotypes, comparable to conventional methods.

Conclusions:

• The proposed attention-based deep network, Read2Pheno, offers a novel approach for microbiome sequence classification and phenotype prediction.
• This method provides biological insights through sequence embeddings and reduces the need for manual interpretation.
• The developed deep learning framework demonstrates strong performance and potential for analyzing microbial community data.