Traitar

Traitar predicts phenotypes from microbial genome sequences to infer metabolic and physiological traits for genome interpretation.

Key Features:

• Automated phenotype prediction: Predicts 67 distinct traits including substrate utilization as carbon and energy sources, oxygen requirements, morphology, antibiotic susceptibility, proteolysis, and enzymatic activities.
• Protein family associations: Identifies protein families linked to specific phenotypes using evolutionary histories and phyletic patterns.
• Machine learning implementation: Uses L1-regularized L2-loss support vector machines to assign phenotypes from protein family data.
• Support for incomplete genomes: Operates on complete and partial genome sequences, including single-cell isolates and simulated drafts.
• Taxonomic breadth: Demonstrated performance across bacterial genomes from 572 species spanning eight phyla.
• Metagenomic application: Applies to metagenomic data to verify and augment manual metabolic reconstructions, exemplified by analysis of two novel Clostridiales species from commercial biogas reactors.

Scientific Applications:

• Microbial trait inference: Infers metabolic and physiological traits that influence ecosystem functioning and human health.
• Large-scale community studies: Enables large-scale analyses of microbial communities by predicting phenotypes from genome sequences, reducing reliance on wet lab experiments.
• Environmental biotechnology: Characterizes microbial capabilities relevant to processes such as wastewater treatment and soil decontamination.
• Metagenomics and novel-species characterization: Assists verification and enhancement of metabolic reconstructions in metagenomic assemblies and novel species descriptions.

Methodology:

Employs L1-regularized L2-loss support vector machines that use phyletic patterns and evolutionary histories of protein families to assign phenotypes from genome-derived protein content.