Treesist-TB

Treesist-TB predicts mutations associated with drug resistance in Mycobacterium tuberculosis (MTB) from whole-genome sequencing data using a customized decision tree approach.

Key Features:

• Customized Decision Tree Approach: Applies a decision tree methodology adapted to TB-specific contexts to reduce overfitting in resistance prediction.
• Genomic Variant Library: Constructs a library of re-occurring genomic variants aggregated across individual studies for genotypic profiling.
• Whole-Genome Sequencing Input: Leverages whole-genome sequencing data to extract and evaluate genomic variants linked to resistance.
• High Predictive Accuracy: Demonstrated accuracies for first-line drugs—rifampicin (RIF) 97.5%, isoniazid (INH) 96.8%, and ethambutol (EMB) 96.8%—compared to TB-Profiler's 97.6%, 96.5%, and 95.8% respectively.
• Enhanced Sensitivity for Second-Line Drugs: Identified additional variants and achieved higher sensitivities for para-aminosalicylic acid (PAS) 64.3% vs 38.8%, cycloserine (CYS) 45.3% vs 30.7%, and ethionamide (ETH) 72.1% vs 71.1% compared to TB-Profiler.
• Overfitting Mitigation: Tailors the model to TB-specific variant patterns to mitigate overfitting common in generic machine learning models.

Scientific Applications:

• Resistance Mutation Prediction: Predicts mutations associated with resistance to first-line and second-line anti-TB drugs from genomic data.
• Variant Discovery: Identifies novel resistance-associated variants not present in existing libraries such as TB-Profiler.
• Genotypic Profiling for Research: Supports research into resistance mechanisms and development of diagnostic markers and treatment strategies for drug-resistant TB.

Methodology:

Uses whole-genome sequencing data to extract and evaluate genomic variants from multiple studies, constructs a library of re-occurring variants, applies a customized decision tree methodology, and compares predictive performance to TB-Profiler.