SUSPECT-PZA

SUSPECT-PZA predicts pyrazinamide resistance by assessing effects of pncA (pyrazinamidase) mutations on protein structure and function using graph-based signatures and machine learning.

Key Features:

• Structure-Guided Prediction: Uses structural information of pncA to assess how variants affect three-dimensional structure, stability, conformation, and molecular interactions.
• Comprehensive Mutation Analysis: Utilizes a curated dataset of 610 pncA mutations with high-confidence experimental and clinical pyrazinamide susceptibility data.
• Graph-Based Signature Methods (mCSM): Employs mCSM graph-based signature methods to computationally evaluate molecular consequences of mutations, including changes in protein stability and conformational dynamics.
• Machine Learning Classifier: Incorporates a machine learning classifier trained on mCSM-derived molecular features that achieves 80% accuracy in predicting pyrazinamide resistance.
• Validation with Clinical Data: Validated against internationally curated clinical datasets with up to 85% accuracy, and screening of 600 Victorian clinical isolates identified previously unreported variants with 71% agreement with drug susceptibility testing.

Scientific Applications:

• Molecular Diagnostics: Predicts pyrazinamide resistance from pncA genotypes to inform molecular diagnostic interpretation.
• Variant Discovery and Interpretation: Identifies and interprets novel pncA variants affecting pyrazinamidase function and potential resistance mechanisms.
• Clinical and Epidemiological Analysis: Supports treatment decision-making and analysis of clinical isolate datasets for surveillance of pyrazinamide resistance.

Methodology:

Curates a dataset of pncA mutations with known clinical outcomes, applies mCSM graph-based signature analyses to predict structural impacts, and trains a machine learning classifier on the resulting molecular features.