Skip to main content
SearchLoginLogin or Signup

Prediction of pyrazinamide resistance in Mycobacterium tuberculosis using structure-based machine learning approaches (lay summary)

This is a lay summary of the article published under the DOI: 10.1101/518142

Published onApr 30, 2023
Prediction of pyrazinamide resistance in Mycobacterium tuberculosis using structure-based machine learning approaches (lay summary)

Scientists used computer models to better understand TB drug resistance

In this study, the researchers gave evidence that computer models can help to predict drug resistance, which is when the target of the drug, like a bacterium or a virus, is no longer effectively killed or controlled by that drug.

They tested their models using one tuberculosis (TB) drug called “pyrazinamide”. They hope this work will help scientists and doctors who are looking for better ways to test for drug resistance.  

TB is caused by bacteria and kills many people around the world. Sometimes the medicines we use to treat TB stop working over time because the bacteria evolves, or changes, to escape the deadly effects of the medicine. 

One such medicine is “pyrazinamide”, which is commonly used to treat TB. Scientists and doctors say it’s important to know if a medicine still works against a disease so that they can better treat their patients. Therefore, the medical community is looking for ways to better test the performance of medicines.  

Not many helpful testing approaches have been discovered so far. In this study researchers attempted to use computer models.

They developed and tested 3 different models. The first model was very simple. The second model was a little bit more complex, and the third model was the most complex.

They used existing information about changes in a gene called ‘pncA’. Gene ‘pncA’ produces an enzyme (a type of protein) through which the medicine pyrazinamide works. 

The researchers developed the models by gradually removing gene changes that did not make TB bacteria resistant to “pyrazinamide”.

They trained the models using strains of TB that are known to be resistant or susceptible to this specific drug.

The researchers found that their models could not correctly identify whether a strain was resistant or not for only 107 (out of 291) samples. They said half of the errors were considered small, where the model could not classify a mutation as either resistant or susceptible. The remaining errors were major, where the models either regarded susceptible strains as resistant, or those resistant as susceptible.

Their most complex model was the best at correctly predicting the highest number (94%) of resistant strains. They however said it would be difficult to learn about resistance using this model. The first and simplest model gave the highest number of major errors, but was best for helping them to understand more about resistance.

When they further tested their best model, the researchers reported that their model correctly predicted 74% resistant isolates, and 54% of the susceptible isolates

The researchers reported that structural changes on the surface of the PncA enzyme were more likely to cause resistance, because 10 out of 11 changes that caused major errors were on the protein surface.

They said major errors could have been due to the fact that when they developed the models, they had assumed mutations would affect the structure of function of gene PncA.

The researchers gave some evidence that computer models might be useful to predict antibiotic resistance for pyrazinamide. They suggest their method may also work for other medicines.

They also said their models can help other scientists to further understand how the resistance actually occurs.

However, the researchers said their model was limited in that it could only tell whether there was a high or low level of resistance to pyrazinamide or not, and not more information about the resistance. They also said they were only able to predict resistance because of changes that occurred in gene pncA, even though TB can be resistant to pyrazinamide because of mutations in other genes.

The researchers recommended that their model could be further improved so that it is more sensitive and specific, and could be used to test in hospitals.

The researchers were based in South Africa and the United Kingdom.


Pyrazinamide is one of four first-line antibiotics used to treat tuberculosis, however antibiotic susceptibility testing for pyrazinamide is problematic. Resistance to pyrazinamide is primarily driven by genetic variation in pncA, an enzyme that converts pyrazinamide into its active form. We curated a derivation dataset of 291 non-redundant, missense amino acid mutations in pncA with associated high-confidence phenotypes from published studies and then trained three different machine learning models to predict pyrazinamide resistance based on sequence- and structure-based features of each missense mutation. The clinical performance of the models was estimated by predicting the binary pyrazinamide resistance phenotype of 2,292 clinical isolates harboring missense mutations in pncA. Overall, this work offers an approach to improve the sensitivity/specificity of pyrazinamide resistance prediction in genetics-based clinical microbiology workflows, highlights novel mutations for future biochemical investigation, and is a proof of concept for using this approach in other drugs such as bedaquiline.


This summary is a free resource intended to make African research and research that affects Africa, more accessible to non-expert global audiences. It was compiled by ScienceLink's team of professional African science communicators as part of the Masakhane MT: Decolonise Science project. ScienceLink has taken every precaution possible during the writing, editing, and fact-checking process to ensure that this summary is easy to read and understand, while accurately reporting on the facts presented in the original research paper. Note, however, that this summary has not been fact-checked or approved by the authors of the original research paper, so this summary should be used as a secondary resource. Therefore, before using, citing or republishing this summary, please verify the information presented with the original authors of the research paper, or email [email protected] for more information.

No comments here
Why not start the discussion?