四川大学学报（自然科学版）2025，Vol.62Issue(3)：548-555,8.DOI:10.19907/j.0490-6756.250003
结合药物语义信息与全基因组的结核分枝杆菌耐药性预测模型
Drug resistance prediction model for mycobacterium tuberculosis by combining drug-related semantic information and whole-genome features
摘要
Abstract
Tuberculosis is one of the main infectious diseases globally,and drug resistance poses a serious threat to public health.Accurate prediction of drug resistance can not only help improve treatment strategies but also reduce the spread of drug-resistant Mycobacterium tuberculosis,thereby enhancing overall public health,especially in resource-limited settings.However,current methods for studying antibiotic resistance typically rely on experimental validation for each drug individually,which is labor-intensive and time-consuming,and fails to meet the demands of long drug development cycles and the limited availability of ex-perimental data.To overcome these challenges,this study introduces a novel drug resistance prediction model that combines whole-genome information with drug-related semantic features to improve the accuracy and generalizability of drug resistance predictions for new drugs.The model employs a hierarchical encoder with an attention mechanism to encode the genomic features of isolates.Subsequently,it employs cross-attention mechanisms to integrate drug information into the feature representation of the isolates,thereby en-hancing the capability of drug resistance prediction.The model was tested using drug-resistant Mycobacte-rium tuberculosis from four first-line drugs and seven second-line drugs,covering representative anti-TB medications commonly used in clinical practice to evaluate its effectiveness.The results show that the model achieves over 80%sensitivity in drug resistance prediction while maintaining high specificity and accuracy.The proposed model reduces the reliance on large-scale testing and provides technical support for rapid screen-ing in resource-limited healthcare settings.关键词
结核分枝杆菌/耐药预测/全基因组/药物语义Key words
Mycobacterium tuberculosis/Drug resistance prediction/Whole genome/Drug semantic分类
信息技术与安全科学引用本文复制引用
赵明明,刘津江,吴畏,孙群,于中华,黄妮妮,陈黎..结合药物语义信息与全基因组的结核分枝杆菌耐药性预测模型[J].四川大学学报（自然科学版）,2025,62(3):548-555,8.基金项目
国家自然科学基金重点项目(62137001) （62137001）
四川省重点研发项目(2023YFG0265) （2023YFG0265）
