Deep learning for predicting 16S rRNA gene copy number.

Area of Science:

• Microbiology
• Bioinformatics
• Computational Biology

Background:

• 16S rRNA gene metabarcoding is crucial for microbiome profiling.
• Accurate 16S rRNA gene copy number (16S GCN) estimation is vital for quantitative microbiome analysis.
• Existing bioinformatic tools for 16S GCN prediction rely on taxonomy or phylogeny.

Purpose of the Study:

• To introduce ANNA16, a novel deep learning approach for estimating 16S GCN directly from 16S gene sequences.
• To evaluate ANNA16's performance against current 16S GCN prediction methods.
• To explore ANNA16's ability to identify informative sequence positions.

Main Methods:

• Developed ANNA16, an Artificial Neural Network Approximator for 16S rRNA gene copy number, utilizing deep learning.
• Trained ANNA16 on 27,579 16S rRNA gene sequences and associated GCN data from the rrnDB database.
• Employed Shapley Additive exPlanations (SHAP) to interpret ANNA16's predictions and identify sequence features.

Main Results:

• ANNA16 demonstrated superior performance in estimating 16S GCN compared to existing algorithms.
• SHAP analysis revealed ANNA16's capacity to identify informative sequence positions without prior phylogenetic knowledge.
• The findings suggest ANNA16's potential for applications beyond 16S GCN prediction.

Conclusions:

• ANNA16 offers a highly accurate, sequence-based method for 16S GCN estimation.
• The deep learning approach provides novel insights into sequence-based GCN prediction.
• ANNA16 has implications for advancing quantitative microbiome research and sequence analysis.