Abstract：
OBJECTIVE: The study aimed to develop a genotypic antimicrobial resistance testing method for Klebsiella pneumoniae using metagenomic sequencing data.
METHODS: We utilized Lasso regression on assembled genomes to identify genetic resistance determinants for six antibiotics (Gentamicin, Tobramycin, Imipenem, Meropenem, Ceftazidime, Trimethoprim/Sulfamethoxazole). The genetic features were weighted, grouped into clusters to establish classifier models. Origin species of detected antibiotic resistant gene (ARG) was determined by novel strategy integrating \"possible species,\" \"gene copy number calculation\" and \"species-specific kmers.\" The performance of the method was evaluated on retrospective case studies.
RESULTS: Our study employed machine learning on 3928 K. pneumoniae isolates, yielding stable models with AUCs > 0.9 for various antibiotics. GenseqAMR, a read-based software, exhibited high accuracy (AUC 0.926-0.956) for short-read datasets. The integration of a species-specific kmer strategy significantly improved ARG-species attribution to an average accuracy of 96.67%. In a retrospective study of 191 K. pneumoniae-positive clinical specimens (0.68-93.39% genome coverage), GenseqAMR predicted 84.23% of AST results on average. It demonstrated 88.76-96.26% accuracy for resistance prediction, offering genotypic AST results with a shorter turnaround time (mean ± SD: 18.34 ± 0.87 hours) than traditional culture-based AST (60.15 ± 21.58 hours). Furthermore, a retrospective clinical case study involving 63 cases showed that GenseqAMR could lead to changes in clinical treatment for 24 (38.10%) cases, with 95.83% (23/24) of these changes deemed beneficial.
CONCLUSIONS: In conclusion, GenseqAMR is a promising tool for quick and accurate AMR prediction in Klebsiella pneumoniae, with the potential to improve patient outcomes through timely adjustments in antibiotic treatment.
METHODS: We utilized Lasso regression on assembled genomes to identify genetic resistance determinants for six antibiotics (Gentamicin, Tobramycin, Imipenem, Meropenem, Ceftazidime, Trimethoprim/Sulfamethoxazole). The genetic features were weighted, grouped into clusters to establish classifier models. Origin species of detected antibiotic resistant gene (ARG) was determined by novel strategy integrating \"possible species,\" \"gene copy number calculation\" and \"species-specific kmers.\" The performance of the method was evaluated on retrospective case studies.
RESULTS: Our study employed machine learning on 3928 K. pneumoniae isolates, yielding stable models with AUCs > 0.9 for various antibiotics. GenseqAMR, a read-based software, exhibited high accuracy (AUC 0.926-0.956) for short-read datasets. The integration of a species-specific kmer strategy significantly improved ARG-species attribution to an average accuracy of 96.67%. In a retrospective study of 191 K. pneumoniae-positive clinical specimens (0.68-93.39% genome coverage), GenseqAMR predicted 84.23% of AST results on average. It demonstrated 88.76-96.26% accuracy for resistance prediction, offering genotypic AST results with a shorter turnaround time (mean ± SD: 18.34 ± 0.87 hours) than traditional culture-based AST (60.15 ± 21.58 hours). Furthermore, a retrospective clinical case study involving 63 cases showed that GenseqAMR could lead to changes in clinical treatment for 24 (38.10%) cases, with 95.83% (23/24) of these changes deemed beneficial.
CONCLUSIONS: In conclusion, GenseqAMR is a promising tool for quick and accurate AMR prediction in Klebsiella pneumoniae, with the potential to improve patient outcomes through timely adjustments in antibiotic treatment.
摘要：
目的：本研究旨在利用宏基因组测序数据建立肺炎克雷伯菌基因型耐药性检测方法。
方法：我们利用组装基因组的Lasso回归来鉴定六种抗生素的遗传抗性决定因素（庆大霉素，妥布霉素,亚胺培南,美罗培南,头孢他啶,甲氧苄啶/磺胺甲恶唑)。遗传特征被加权，分组为集群以建立分类器模型。通过整合“可能物种”的新策略确定检测到的抗生素抗性基因（ARG）的起源物种，“基因拷贝数计算”和“物种特异性kmers”。在回顾性案例研究中评估了该方法的性能。
结果：我们的研究对3928株肺炎克雷伯菌进行了机器学习，对于各种抗生素，产生AUC>0.9的稳定模型。GenseqAMR,基于读取的软件，短读数数据集表现出高精度(AUC0.926-0.956)。物种特异性kmer策略的整合显着提高了ARG物种归因，平均准确率为96.67％。在191个肺炎克雷伯菌阳性临床标本的回顾性研究中（基因组覆盖率为0.68％-93.39％），GenseqAMR平均预测了84.23%的AST结果。它显示了88.76%-96.26%的电阻预测精度，提供基因型AST结果的周转时间（平均±SD：18.34±0.87小时）比基于传统培养的AST（60.15±21.58小时）短。此外，一项涉及63例的回顾性临床病例研究表明，GenseqAMR可导致24例（38.10％）患者的临床治疗发生变化，95.83%(23/24)的这些变化被认为是有益的。
结论：结论：GenseqAMR是一种用于快速准确预测肺炎克雷伯菌AMR的有前途的工具，通过及时调整抗生素治疗，有可能改善患者的预后。
方法：我们利用组装基因组的Lasso回归来鉴定六种抗生素的遗传抗性决定因素（庆大霉素，妥布霉素,亚胺培南,美罗培南,头孢他啶,甲氧苄啶/磺胺甲恶唑)。遗传特征被加权，分组为集群以建立分类器模型。通过整合“可能物种”的新策略确定检测到的抗生素抗性基因（ARG）的起源物种，“基因拷贝数计算”和“物种特异性kmers”。在回顾性案例研究中评估了该方法的性能。
结果：我们的研究对3928株肺炎克雷伯菌进行了机器学习，对于各种抗生素，产生AUC>0.9的稳定模型。GenseqAMR,基于读取的软件，短读数数据集表现出高精度(AUC0.926-0.956)。物种特异性kmer策略的整合显着提高了ARG物种归因，平均准确率为96.67％。在191个肺炎克雷伯菌阳性临床标本的回顾性研究中（基因组覆盖率为0.68％-93.39％），GenseqAMR平均预测了84.23%的AST结果。它显示了88.76%-96.26%的电阻预测精度，提供基因型AST结果的周转时间（平均±SD：18.34±0.87小时）比基于传统培养的AST（60.15±21.58小时）短。此外，一项涉及63例的回顾性临床病例研究表明，GenseqAMR可导致24例（38.10％）患者的临床治疗发生变化，95.83%(23/24)的这些变化被认为是有益的。
结论：结论：GenseqAMR是一种用于快速准确预测肺炎克雷伯菌AMR的有前途的工具，通过及时调整抗生素治疗，有可能改善患者的预后。
