GenTB

GenTB predicts antibiotic resistance phenotypes in Mycobacterium tuberculosis from next-generation sequencing data using machine-learning models for genomic resistance profiling.

Key Features:

• Machine Learning Algorithms: Implements a Random Forest (RF) classifier that predicts phenotypic resistance to 13 anti-tuberculosis drugs and a Wide and Deep Neural Network (WDNN) that predicts resistance to 10 anti-tuberculosis drugs.
• Benchmarking Performance: In comparative evaluations, GenTB-RF achieved mean sensitivity 77.6% (95% CI 76.6–78.5%) and GenTB-WDNN 75.4% (95% CI 74.5–76.4%), compared with TB-Profiler 74.4% and Mykrobe 71.9%, with a marginal trade-off in specificity while maintaining high accuracy.
• Geographic and Mutation Data Mapping: Provides mapping of geographic resistance patterns and mutation data to support epidemiological analyses of resistant M. tuberculosis strains.
• Sequencing Input and Quality Considerations: Analyzes next-generation sequencing-derived M. tuberculosis DNA sequences and notes that genotypic resistance sensitivity is substantially reduced for isolates with low sequencing depth (<10x across 95% of the genome).

Scientific Applications:

• Clinical diagnostics: Enables rapid genotype-based prediction of drug resistance to inform selection of anti-tuberculosis treatment regimens.
• Public health surveillance: Supports monitoring and mapping of drug-resistant TB prevalence and geographic spread.
• Epidemiological research: Facilitates analysis of mutation prevalence and the evolution and dissemination of multidrug-resistant M. tuberculosis.

Methodology:

Analyzes M. tuberculosis DNA sequences from next-generation sequencing data using a Random Forest classifier and a Wide and Deep Neural Network to predict resistance phenotypes, and incorporates sequence quality control considerations including reduced sensitivity for isolates with <10x depth across 95% of the genome.