Demixer

Demixer delineates different strains of a microbial species in mixed infection samples using whole genome sequencing (WGS) data by combining reference-based and reference-free techniques to identify de novo strains and quantify known strains.

Key Features:

• Probabilistic generative model: Extends a topic model widely used in text mining to represent known mutations and discover novel ones.
• Reference-based and reference-free combination: Identifies de novo strains not present in reference databases while quantifying proportions of known strains.
• Scalable processing: Parallelization and heuristics enable processing of large datasets such as CRyPTIC (Comprehensive Resistance Prediction for Tuberculosis: an International Consortium).
• Strain detection and quantification: Detects identity and proportions of mixed strains with high accuracy in both synthetic and experimental benchmark datasets.
• Generalizability: Applicable to any bacterial and viral WGS data.

Scientific Applications:

• Tuberculosis mixed infection analysis: Reveals novel high confidence mixed infections and provides insights into global frequency of mixed infection and its association to drug resistance.
• Multi-drug resistant and hetero-resistant TB characterization: Delineates multiple strains of TB-causing bacterium with varying degrees of drug susceptibility present in an individual.
• Large-scale epidemiological studies: Processes large consortium datasets such as CRyPTIC for population-level mixed infection analysis.

Methodology:

Demixer extends a topic model widely used in text mining to represent known mutations and discover novel ones, with parallelization and heuristics to enable large dataset processing.