MetaCONNET: A metagenomic polishing tool for long-read assemblies

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• High-quality genome assemblies are crucial for metagenomic studies.
• Long-read sequencing offers advantages for metagenome assembly but has high error rates.
• Existing polishing tools often lack scalability for complex metagenomic datasets.

Purpose of the Study:

• To introduce MetaCONNET, a novel deep-learning based polishing tool for metagenomic assemblies.
• To address the limitations of current tools in handling the complexity and uneven sequencing depth of metagenomic data.
• To evaluate MetaCONNET's performance against established polishing tools.

Main Methods:

• Development of a deep-learning model for polishing metagenomic sequences.
• Comparative analysis of MetaCONNET against Medaka, CONNET, and NextPolish.
• Evaluation metrics included accuracy, coverage, contiguity, and computational resource consumption.

Main Results:

• MetaCONNET demonstrated superior performance in polishing metagenomic assemblies.
• The tool effectively improved accuracy and coverage compared to existing methods.
• MetaCONNET showed competitive or improved contiguity and resource efficiency.

Conclusions:

• MetaCONNET is a valuable and effective tool for polishing metagenomic assemblies.
• The deep-learning approach enhances the accuracy of genome reconstruction from complex samples.
• MetaCONNET has broad applicability for diverse metagenomic research.