Deep learning for Metagenome Assembly Error Detection (DeepMAsED)

Deep learning for Metagenome Assembly Error Detection (DeepMAsED) detects misassembled contigs in metagenomic assemblies using deep learning to evaluate assembly quality without requiring reference genomes.

Key Features:

• Reference-free misassembly detection: Identifies misassembled contigs in metagenomic assemblies without relying on closely related reference genomes.
• Deep learning-based classification: Utilizes deep learning models to analyze contig features and predict misassemblies.
• High accuracy: Demonstrates superior accuracy compared to state-of-the-art methods on large and complex metagenome assemblies.
• In silico training pipeline: Provides a pipeline for generating realistic, large-scale metagenome assemblies for model training and testing.
• Broad taxonomic applicability: Capable of detecting misassemblies across a wide diversity of bacteria and archaea.
• Misassembly rate estimation: Estimates misassembly rates in metagenome assembly datasets with high precision.

Scientific Applications:

• Microbial ecology: Evaluates assembly quality to improve accuracy of community composition and functional inference.
• Environmental microbiology: Assesses metagenome assemblies from environmental samples to ensure reliable downstream analyses.
• Evolutionary biology: Detects assembly errors that could bias comparative and evolutionary inferences.

Methodology:

Performs reference-free contig analysis using deep learning and an in silico pipeline that generates realistic large-scale metagenome assemblies for model training and testing, and outputs misassembly predictions and misassembly rate estimates.