DeepMobilome: predicting mobile genetic elements using sequencing reads of microbiomes

Area of Science:

• Microbiology
• Genomics
• Bioinformatics

Background:

• Mobile genetic elements (MGEs) are key drivers of antibiotic resistance gene (ARG) acquisition in microbial communities.
• Analyzing the ARG-carrying mobilome is crucial for understanding antibiotic resistance evolution.
• Existing MGE prediction tools struggle with metagenomic data, leading to high false positive rates.

Purpose of the Study:

• To develop a robust method for accurately identifying MGEs within complex microbiomes.
• To overcome the limitations of current MGE prediction tools in metagenomic analysis.
• To enhance the understanding of ARG dissemination and support interventions against antibiotic resistance.

Main Methods:

• Developed DeepMobilome, a novel approach utilizing a convolutional neural network.
• Trained the model on read alignment data from sequence alignment map (SAM) files.
• Validated DeepMobilome on simulated and real microbiome datasets, comparing performance against existing methods.

Main Results:

• DeepMobilome achieved a high validation accuracy of 0.99.
• Outperformed MGEfinder (F1=0.755) and ISMapper (F1=0.670) in single-genome tests with an F1-score of 0.935.
• Successfully identified six ARG-carrying MGEs in real microbiome data, demonstrating robustness and reliability.

Conclusions:

• DeepMobilome provides a significant advancement in identifying ARG-carrying MGEs from metagenomic data.
• The tool offers improved accuracy and reliability compared to existing methods.
• DeepMobilome is a valuable asset for studying ARG dissemination and developing strategies to combat antibiotic resistance.