GeneMark-HM

GeneMark-HM predicts protein-coding genes in assembled human metagenomic sequences to improve annotation of microbiome-derived prokaryotic genomes.

Key Features:

• Algorithm integration: Integrates MetaGeneMark-2, GeneMarkS-2, and Genemark.hmm-2 into a unified workflow for metagenomic gene prediction.
• Multi-algorithm modeling: Uses three distinct algorithm types to generate nearly optimal models of protein-coding regions from either pre-computed libraries or de novo construction.
• Genome reconstruction: Reconstructs nearly complete prokaryotic genomes from metagenomic reads to support downstream gene prediction.
• Pan-genome database construction: Builds species-level pan-genome model libraries for the human microbiome using the self-training GeneMarkS-2 algorithm.
• Similarity-based model selection: Uses initial contig gene predictions as queries for a rapid similarity search against the pan-genome database and selects best matches to guide model choice.
• GC-composition models: Applies specialized models tailored for sequences with specific GC compositions for contigs that cannot be assigned to pan-genomes.
• Model selection and annotation workflow: Executes a dedicated workflow that performs model selection and subsequent gene annotation.
• Empirical performance: Demonstrated improved gene annotation on simulated metagenomes compared to current state-of-the-art tools.

Scientific Applications:

• Gene prediction in human metagenomes: Accurate annotation of protein-coding genes in assembled sequences derived from the human microbiome.
• Identification of uncultured species: Facilitates identification of thousands of new uncultured candidate bacterial species via reconstructed prokaryotic genomes and sequenced isolates.
• Pan-genome analysis: Enables construction and use of species-level pan-genomes for the human microbiome.
• Annotation of novel sequences: Enhances gene prediction accuracy in novel metagenomic sequences using models derived from reconstructed genomes and isolates.
• Benchmarking and validation: Provides a framework for comparing gene annotation performance on simulated metagenomes against other methods.

Methodology:

Integrates MetaGeneMark-2, GeneMarkS-2, and Genemark.hmm-2; reconstructs nearly complete prokaryotic genomes from metagenomic reads; uses GeneMarkS-2 self-training to build pan-genome model libraries; performs initial gene predictions per contig which are used as queries in a rapid similarity search against the pan-genome database to select best models; sources models from pre-computed libraries or constructs them de novo; applies GC-composition-specific models for unassigned contigs.