Genome medicine · 2025
Advantages of updated WHO mutation catalog combined with existing whole-genome sequencing-based approaches for Mycobacterium tuberculosis resistance prediction.
СтатьяПоддержка исследования: вне правительства СШАНазвание и аннотация: Перевод в очереди
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Chen Yiwang, Zhang Xuecong, Liang Jialei, Jiang Qi, Peierdun Mijiti, Xu Peng, Takiff Howard E, Gao Qian
Аннотация
BACKGROUND: The WHO recently released a second edition of the mutation catalog for predicting drug resistance in Mycobacterium tuberculosis (MTB). This study evaluated its effectiveness compared to existing whole-genome sequencing (WGS)-based prediction methods and proposes a novel approach for its optimization.
METHODS: We tested the accuracy of five tools-the WHO catalog, TB Profiler, SAM-TB, GenTB, and MD-CNN-for predicting drug susceptibility on a global dataset of 36,385 MTB isolates with high-quality phenotypic drug susceptibility testing (DST) and WGS data. By integrating the genotypic DST predictions of these five tools in an ensemble machine learning framework, we developed an improved computational model for MTB drug susceptibility prediction. We then validated the ensemble model on 860 MTB isolates with phenotypic and WGS data collected in Shenzhen, China (2013-2019) and Valencia, Spain (2014-2016).
RESULTS: Among the five genotypic DST tools for predicting susceptibility to ten drugs, MD-CNN exhibited the highest overall performance (AUC 92.1%; 95% CI 89.8-94.4%). The WHO catalog demonstrated the highest specificity of 97.3% (95% CI 95.8-98.4%), while TB Profiler had the best sensitivity at 79.5% (95% CI 71.8-86.2%). The ensemble machine learning model (AUC 93.4%; 95% CI 91.4-95.4%) outperformed all of the five individual tools, with a specificity of 95.4% (95% CI 93.0-97.6%) and a sensitivity of 84.1% (95% CI 78.8-88.8%), principally due to considerable improvements in second-line drug resistance predictions (AUC 91.8%; 95% CI 89.6-94.0%).
CONCLUSIONS: The second edition of the WHO MTB mutation catalog does not, by itself, perform better than existing tools for predicting MTB drug resistance. An integrative approach combining the WHO catalog with other genotypic DST methods significantly enhances prediction accuracy.
METHODS: We tested the accuracy of five tools-the WHO catalog, TB Profiler, SAM-TB, GenTB, and MD-CNN-for predicting drug susceptibility on a global dataset of 36,385 MTB isolates with high-quality phenotypic drug susceptibility testing (DST) and WGS data. By integrating the genotypic DST predictions of these five tools in an ensemble machine learning framework, we developed an improved computational model for MTB drug susceptibility prediction. We then validated the ensemble model on 860 MTB isolates with phenotypic and WGS data collected in Shenzhen, China (2013-2019) and Valencia, Spain (2014-2016).
RESULTS: Among the five genotypic DST tools for predicting susceptibility to ten drugs, MD-CNN exhibited the highest overall performance (AUC 92.1%; 95% CI 89.8-94.4%). The WHO catalog demonstrated the highest specificity of 97.3% (95% CI 95.8-98.4%), while TB Profiler had the best sensitivity at 79.5% (95% CI 71.8-86.2%). The ensemble machine learning model (AUC 93.4%; 95% CI 91.4-95.4%) outperformed all of the five individual tools, with a specificity of 95.4% (95% CI 93.0-97.6%) and a sensitivity of 84.1% (95% CI 78.8-88.8%), principally due to considerable improvements in second-line drug resistance predictions (AUC 91.8%; 95% CI 89.6-94.0%).
CONCLUSIONS: The second edition of the WHO MTB mutation catalog does not, by itself, perform better than existing tools for predicting MTB drug resistance. An integrative approach combining the WHO catalog with other genotypic DST methods significantly enhances prediction accuracy.