MetaGUN

MetaGUN predicts protein-coding genes in metagenomic fragments by combining k-mer based phylogenetic binning, Support Vector Machine (SVM) classification using codon usage and TIS features, and translation initiation site refinement to improve gene boundary identification in complex microbial communities.

Key Features:

• Three-Stage Prediction Strategy: Divides prediction into phylogenetic group classification, protein-coding sequence identification, and translation initiation site adjustment.
• Phylogenetic Group Classification (k-mer based sequence binning): Classifies input metagenomic fragments into phylogenetic groups using k-mer based sequence binning.
• Protein-Coding Sequence Identification (SVM classifiers): Employs Support Vector Machine classifiers that integrate entropy density profiles (EDP) of codon usage, translation initiation site (TIS) scores, and open reading frame (ORF) length.
• Translation Initiation Site Adjustment (modified MetaTISA): Refines predicted TISs using a modified version of MetaTISA to improve start-site determination.
• Universal Module: Uses a set of representative species to identify known protein-coding sequences.
• Novel Module: Detects potential functional DNA sequences with conserved domains to identify candidate novel genes.

Scientific Applications:

• Gene boundary prediction: Improves prediction of both 3' and 5' gene ends across fragments of varying lengths.
• Benchmarking against metagenomic gene finders: Demonstrated superior performance compared to existing metagenomic gene finders in the reported evaluations.
• Human gut microbiome analysis: Applied to human gut microbiome samples where it identified thousands of additional genes with supporting evidence.
• Novel gene discovery: Predicts a higher number of potential novel genes, aiding exploration of microbial diversity and new genetic functions.

Methodology:

Computational steps explicitly include k-mer based sequence binning for phylogenetic classification; SVM classifiers integrating entropy density profiles (EDP) of codon usage, TIS scores, and ORF length for protein-coding identification; TIS refinement using a modified MetaTISA; a universal module based on representative species; and a novel module for detecting conserved-domain-containing sequences.