What is it about?
Tuberculosis remains a major global health threat, and one of the main obstacles to its treatment is the emergence of drug-resistant strains. A key gene involved in this resistance is pncA, which encodes an enzyme responsible for activating pyrazinamide, a frontline antibiotic. When this gene carries mutations, the drug stops working. To better understand how these mutations affect the shape of the protein — and potentially design better treatments — researchers need computational tools to predict three-dimensional protein structures. In this study, we systematically compared eight such tools, ranging from classical template-based methods to modern AI-driven approaches, by modeling 204 clinically relevant mutations in the PZase protein. The resulting structures were assessed using five independent quality metrics to identify which tools perform best for this type of analysis.
Featured Image
Photo by National Institute of Allergy and Infectious Diseases on Unsplash
Why is it important?
Most benchmarks of protein structure prediction focus on normal, unaltered proteins. Very few evaluate how well these tools handle single-point mutations, tiny but clinically critical changes in a protein's sequence. Our work fills this gap by providing the first large-scale, automated comparison of eight modeling tools specifically applied to mutation modeling in a drug-resistance context. We also developed and made publicly available an automated pipeline that handles both structure prediction and validation, lowering the barrier for research groups with limited computational resources. The findings clearly show that modern deep learning-based tools, particularly AlphaFold3 and trRosetta, outperform traditional methods in structural quality, and can serve as a reliable foundation for future studies on drug binding and resistance mechanisms in tuberculosis.
Perspectives
This project grew from a genuine need we observed in our research group: there was no clear, practical guidance on which computational tool to choose when modeling mutant proteins for drug-resistance studies. Working across disciplines, combining biomedical sciences and computer science, allowed us to tackle this problem systematically. Beyond tuberculosis, we believe the automated pipeline we developed here can be adapted to other pathogens and resistance-related genes, making structural bioinformatics more accessible to the broader scientific community. We hope this work helps researchers make more informed methodological choices and contributes, even modestly, to the global effort against antimicrobial resistance.
Veridiana Piva Richter
Universidade Federal do Rio Grande
Read the Original
This page is a summary of: Comparative Analysis of Protein Structure Prediction Tools for Modeling Single-Point Mutations in the pncA gene of Mycobacterium tuberculosis, March 2026, ACM (Association for Computing Machinery),
DOI: 10.1145/3748522.3779838.
You can read the full text:
Contributors
The following have contributed to this page
