DeepMAP: Deep CNN Classifiers Applied to Optical Mapping for Fast and Precise Species-Level Metagenomic Analysis.

Area of Science:

• Genomics and Bioinformatics
• Computational Biology
• Microbiome Analysis

Background:

• DNA optical mapping is valuable for structural variant calling and genome assembly verification.
• Current alignment algorithms limit optical mapping's application in complex settings like microbiome species identification due to speed and database size constraints.

Purpose of the Study:

• To develop a novel genomic classification pipeline for fast and accurate species identification using DNA optical mapping data.
• To overcome existing algorithmic limitations for high-throughput microbiome analysis.

Main Methods:

• Development of DeepMAP, a pipeline employing deep convolutional neural networks for optical mapping data analysis.
• Evaluation of DeepMAP on genomic DNA from bacterial mixtures to assess its classification accuracy and speed.

Main Results:

• DeepMAP achieves fast and accurate assignment of optical maps to their respective genomes.
• The pipeline demonstrates superior performance with evolutionary divergent sequences, showing robustness to unknown strains in metagenomic samples.
• Species-level resolution was reached with a true positive rate of ~75% and a false positive rate <1%, significantly outperforming existing methods.

Conclusions:

• DeepMAP represents a significant advancement in applying DNA optical mapping to complex microbiome analyses.
• The pipeline's speed and accuracy enable efficient species identification, broadening the utility of optical mapping technologies.
• DeepMAP's robustness to sequence divergence makes it suitable for real-world metagenomic applications.