Drug resistance prediction for Mycobacterium tuberculosis with reference graphs

Area of Science:

• Genomics
• Microbiology
• Computational Biology

Background:

• Tuberculosis (TB) presents a global health challenge, exacerbated by increasing antimicrobial resistance.
• The World Health Organization (WHO) aims for universal access to drug susceptibility testing (DST).
• Current phenotypic DST methods are slow and infrastructure-intensive, necessitating faster alternatives.

Purpose of the Study:

• To develop and evaluate DrPRG (Drug resistance Prediction with Reference Graphs), a novel genotypic DST tool for *Mycobacterium tuberculosis*.
• To assess DrPRG's performance against existing tools like Mykrobe and TBProfiler.
• To identify novel resistance-conferring variations in *M. tuberculosis*.

Main Methods:

• Construction of a *Mycobacterium tuberculosis* drug resistance reference graph using the Pandora bacterial reference graph method.
• Incorporation of a global dataset of isolates with diverse drug susceptibility profiles.
• Benchmarking DrPRG against Mykrobe and TBProfiler using 44,709 Illumina and 138 Nanopore samples.

Main Results:

• DrPRG demonstrated significantly improved sensitivity and specificity for certain drugs compared to existing tools.
• DrPRG exhibited reduced computational memory usage and faster runtimes than Mykrobe and TBProfiler.
• Novel insights into resistance-conferring variations, including gene deletions (*katG*, *pncA*), were identified.

Conclusions:

• DrPRG offers a more accurate and efficient approach to genotypic DST for tuberculosis.
• The tool's performance and resource efficiency support the WHO's goal of accessible DST.
• Further investigation into identified mutations may refine TB resistance classification.