Systems Genomics Modeling of Multi-drug Resistance in Mycobacterium tuberculosis

Systems Genomics Modelling of Drug Resistance in Mycobacterium tuberculosis 

Problem: 

Multi-drug resistance presents a growing challenge to global tuberculosis (TB) control efforts, with a particularly pressing concern in developing nations where robust disease surveillance and monitoring are lacking. Nepal is one of the countries with a high TB burden. TB is caused by an airborne bacterium, Mycobacterium tuberculosis (Mtb). Treatment for TB involves antibiotics, however, some Mtb becomes resistant to these drugs, forming drug-resistant TB, which is rapidly increasing and poses a major health threat. The emergence of drug resistance TB is linked to specific mutations in Mtb genome. However, the genetic foundations of drug resistance in TB are not fully understood. The existing catalog of well-characterized Mtb drug-resistant mutations falls short in elucidating numerous instances of drug-resistant TB. Identifying new drug-resistant TB mutations is challenging given the genetic diversity of TB and the complex mechanisms driving the drug resistance.  

Research Aim: 

To address these challenges, here, we aim to develop a machine learning method to predict drug-resistant TB leveraging a large pool of TB genomic data. Additionally, we will explore the underlying metabolic adaptation in drug-resistant TB using genome-scale metabolic modeling. 

Current stage of our research: 

We first developed a massively scalable whole genome sequencing (WGS) data processing pipeline capable of running on a personal computer as well as a cluster of computation units in a distributed computing environment. The pipeline includes mapping WGS reads to a pan-susceptible reference Mtb genome, predicting mutations using three different variant calling tools, and polling among their results to robustly identify mutations in each Mtb isolate. Using this pipeline, we analyzed over 5000 publicly available Mtb WGS from clinical isolates and identified mutations in the Mtb genomes. Further, we have performed an extensive evaluation of some of the existing state-of-art deep-learning methods that predict drug-resistant TB. 

Outcomes so far: 

Our preliminary evaluation of some of the existing state-of-art deep-learning methods for predicting drug-resistant TB heavily reliant on the mutational status of a limited set of well-studied Mtb genes. These methods lack the capability to discover new genes linked to drug-resistant TB. We are currently exploring various strategies to enhance our predictive methods and identify novel genes associated with drug-resistant TB. 

Funding agencies: 

Royal Society of Tropical Medicine and Hygiene (RSTMH) Small Grants 2021 

Team members: 

NAAMII: Parikshit Prasai; Bishesh Khanal, PhD; Raunak Shrestha, PhD 

Collaborators: Dipali Singh, PhD (Quadram Institute, UK)