DeepMicroClass sorts metagenomic contigs into prokaryotes, eukaryotes and viruses

Area of Science:

• Microbiology
• Bioinformatics
• Genomics

Background:

• Metagenomic sequence classification is crucial for understanding microbial communities.
• Existing binary classifiers are limited due to complex, multi-source environmental DNA.
• Accurate classification of diverse sequence types is needed.

Purpose of the Study:

• Introduce DeepMicroClass, a deep-learning classifier for five metagenomic sequence types.
• Evaluate DeepMicroClass performance against state-of-the-art methods.
• Demonstrate the utility of DeepMicroClass in ecological case studies.

Main Methods:

• Developed a deep-learning model for sequence classification.
• Classified contigs into viruses (prokaryotic/eukaryotic hosts), chromosomes (eukaryotic/prokaryotic), and plasmids.
• Benchmarked performance on synthetic and real-world metagenomic datasets (CAMI II, coastal time-series, HOT/BATS).

Main Results:

• DeepMicroClass demonstrated high performance across various sequence lengths (500 bp–100 kbps).
• Outperformed other predictors in classifying eukaryotic, plasmid, and viral contigs on synthetic data.
• Showed comparable performance to geNomad and VirSorter2 for viral classification on the CAMI II marine dataset.
• Identified microbial eukaryotes and prokaryotic viruses as integral community components in a coastal study.
• Observed seasonal viral abundance patterns in subsurface waters at HOT and BATS.

Conclusions:

• DeepMicroClass offers a powerful, accurate tool for metagenomic sequence classification.
• The tool enhances the study of under-appreciated sequence types like viruses and plasmids.
• Findings highlight the ecological significance of viruses and microbial eukaryotes in marine environments.