NC｜K-MARVEL基于蛋白k-mer的抗微生物药物耐药性虚拟探索实验室

摘要

抗微生物药物耐药性（antimicrobial resistance, AMR）的快速扩散亟需新型计算监测工具。现有二代测序数据分析方法存在权衡：基于组装的方法计算繁重，而长读长直接比对受高错误率影响，难以识别赋予耐药的点突变。作者提出K-MARVEL（K-Mer based Antimicrobial Resistance Virtual Exploration Lab），一种开源方法，可从短读与长读数据集中捕获抗药性基因（antimicrobial resistance gene, ARG）及赋予耐药的突变。该方法在蛋白k-mer空间运行，可容忍核苷酸水平测序错误。作者在22个菌种、209个长读与205个短读数据集上基准测试，获得短读F1=0.976、长读F1=0.958，在速度与内存上优于基于组装的方法；在同源模型（homology model）ARG上，K-MARVEL对7个常用短读分类器F1达0.979、对2个常用长读分类器F1达0.961。K-MARVEL可直接从原始测序数据准确识别同源ARG以及携带耐药突变（含多种变体、变异模型与过表达模型）的结构基因。

keyword：抗微生物药物耐药性（antimicrobial resistance, AMR）；抗药性基因（antimicrobial resistance gene, ARG）；蛋白k-mer（protein k-mer）；牛津纳米孔长读长（Oxford Nanopore, ONT）；同源模型（homology model）；变异模型（variant model）；过表达模型（overexpression model）；宏基因组（metagenome）

Mahar N. S., Branders S., Grabherr M. G., Gupta I., Ahmad R. (2026). K-MARVEL: K-Mer-based antimicrobial resistance virtual exploration lab. Nature Communications. https://doi.org/10.1038/s41467-026-77438-8
期刊：Nature Communications（2025年度JCR影响因子IF=18.1，五年IF=18.9，JCR Q1，多学科科学；ISSN 2041-1723，Nature Portfolio）
收稿：2026年4月17日；接受：2026年8月17日；在线发布：2026年9月3日

研究总结

#mermaid-svg-FHjTkKN7YVsWmGwP{font-family:"trebuchet ms",verdana,arial,sans-serif;font-size:16px;fill:#333;}@keyframes edge-animation-frame{from{stroke-dashoffset:0;}}@keyframes dash{to{stroke-dashoffset:0;}}#mermaid-svg-FHjTkKN7YVsWmGwP .edge-animation-slow{stroke-dasharray:9,5!important;stroke-dashoffset:900;animation:dash 50s linear infinite;stroke-linecap:round;}#mermaid-svg-FHjTkKN7YVsWmGwP .edge-animation-fast{stroke-dasharray:9,5!important;stroke-dashoffset:900;animation:dash 20s linear infinite;stroke-linecap:round;}#mermaid-svg-FHjTkKN7YVsWmGwP .error-icon{fill:#552222;}#mermaid-svg-FHjTkKN7YVsWmGwP .error-text{fill:#552222;stroke:#552222;}#mermaid-svg-FHjTkKN7YVsWmGwP .edge-thickness-normal{stroke-width:1px;}#mermaid-svg-FHjTkKN7YVsWmGwP .edge-thickness-thick{stroke-width:3.5px;}#mermaid-svg-FHjTkKN7YVsWmGwP .edge-pattern-solid{stroke-dasharray:0;}#mermaid-svg-FHjTkKN7YVsWmGwP .edge-thickness-invisible{stroke-width:0;fill:none;}#mermaid-svg-FHjTkKN7YVsWmGwP .edge-pattern-dashed{stroke-dasharray:3;}#mermaid-svg-FHjTkKN7YVsWmGwP .edge-pattern-dotted{stroke-dasharray:2;}#mermaid-svg-FHjTkKN7YVsWmGwP .marker{fill:#333333;stroke:#333333;}#mermaid-svg-FHjTkKN7YVsWmGwP .marker.cross{stroke:#333333;}#mermaid-svg-FHjTkKN7YVsWmGwP svg{font-family:"trebuchet ms",verdana,arial,sans-serif;font-size:16px;}#mermaid-svg-FHjTkKN7YVsWmGwP p{margin:0;}#mermaid-svg-FHjTkKN7YVsWmGwP .edge{stroke-width:3;}#mermaid-svg-FHjTkKN7YVsWmGwP .section--1 rect,#mermaid-svg-FHjTkKN7YVsWmGwP .section--1 path,#mermaid-svg-FHjTkKN7YVsWmGwP .section--1 circle,#mermaid-svg-FHjTkKN7YVsWmGwP .section--1 polygon,#mermaid-svg-FHjTkKN7YVsWmGwP .section--1 path{fill:hsl(240, 100%, 76.2745098039%);}#mermaid-svg-FHjTkKN7YVsWmGwP .section--1 text{fill:#ffffff;}#mermaid-svg-FHjTkKN7YVsWmGwP .node-icon--1{font-size:40px;color:#ffffff;}#mermaid-svg-FHjTkKN7YVsWmGwP .section-edge--1{stroke:hsl(240, 100%, 76.2745098039%);}#mermaid-svg-FHjTkKN7YVsWmGwP .edge-depth--1{stroke-width:17;}#mermaid-svg-FHjTkKN7YVsWmGwP .section--1 line{stroke:hsl(60, 100%, 86.2745098039%);stroke-width:3;}#mermaid-svg-FHjTkKN7YVsWmGwP .disabled,#mermaid-svg-FHjTkKN7YVsWmGwP .disabled circle,#mermaid-svg-FHjTkKN7YVsWmGwP .disabled text{fill:lightgray;}#mermaid-svg-FHjTkKN7YVsWmGwP .disabled text{fill:#efefef;}#mermaid-svg-FHjTkKN7YVsWmGwP .section-0 rect,#mermaid-svg-FHjTkKN7YVsWmGwP .section-0 path,#mermaid-svg-FHjTkKN7YVsWmGwP .section-0 circle,#mermaid-svg-FHjTkKN7YVsWmGwP .section-0 polygon,#mermaid-svg-FHjTkKN7YVsWmGwP .section-0 path{fill:hsl(60, 100%, 73.5294117647%);}#mermaid-svg-FHjTkKN7YVsWmGwP .section-0 text{fill:black;}#mermaid-svg-FHjTkKN7YVsWmGwP .node-icon-0{font-size:40px;color:black;}#mermaid-svg-FHjTkKN7YVsWmGwP .section-edge-0{stroke:hsl(60, 100%, 73.5294117647%);}#mermaid-svg-FHjTkKN7YVsWmGwP .edge-depth-0{stroke-width:14;}#mermaid-svg-FHjTkKN7YVsWmGwP .section-0 line{stroke:hsl(240, 100%, 83.5294117647%);stroke-width:3;}#mermaid-svg-FHjTkKN7YVsWmGwP .disabled,#mermaid-svg-FHjTkKN7YVsWmGwP .disabled circle,#mermaid-svg-FHjTkKN7YVsWmGwP .disabled text{fill:lightgray;}#mermaid-svg-FHjTkKN7YVsWmGwP .disabled text{fill:#efefef;}#mermaid-svg-FHjTkKN7YVsWmGwP .section-1 rect,#mermaid-svg-FHjTkKN7YVsWmGwP .section-1 path,#mermaid-svg-FHjTkKN7YVsWmGwP .section-1 circle,#mermaid-svg-FHjTkKN7YVsWmGwP .section-1 polygon,#mermaid-svg-FHjTkKN7YVsWmGwP .section-1 path{fill:hsl(80, 100%, 76.2745098039%);}#mermaid-svg-FHjTkKN7YVsWmGwP .section-1 text{fill:black;}#mermaid-svg-FHjTkKN7YVsWmGwP .node-icon-1{font-size:40px;color:black;}#mermaid-svg-FHjTkKN7YVsWmGwP .section-edge-1{stroke:hsl(80, 100%, 76.2745098039%);}#mermaid-svg-FHjTkKN7YVsWmGwP .edge-depth-1{stroke-width:11;}#mermaid-svg-FHjTkKN7YVsWmGwP .section-1 line{stroke:hsl(260, 100%, 86.2745098039%);stroke-width:3;}#mermaid-svg-FHjTkKN7YVsWmGwP .disabled,#mermaid-svg-FHjTkKN7YVsWmGwP .disabled circle,#mermaid-svg-FHjTkKN7YVsWmGwP .disabled text{fill:lightgray;}#mermaid-svg-FHjTkKN7YVsWmGwP .disabled text{fill:#efefef;}#mermaid-svg-FHjTkKN7YVsWmGwP .section-2 rect,#mermaid-svg-FHjTkKN7YVsWmGwP .section-2 path,#mermaid-svg-FHjTkKN7YVsWmGwP .section-2 circle,#mermaid-svg-FHjTkKN7YVsWmGwP .section-2 polygon,#mermaid-svg-FHjTkKN7YVsWmGwP .section-2 path{fill:hsl(270, 100%, 76.2745098039%);}#mermaid-svg-FHjTkKN7YVsWmGwP .section-2 text{fill:#ffffff;}#mermaid-svg-FHjTkKN7YVsWmGwP .node-icon-2{font-size:40px;color:#ffffff;}#mermaid-svg-FHjTkKN7YVsWmGwP .section-edge-2{stroke:hsl(270, 100%, 76.2745098039%);}#mermaid-svg-FHjTkKN7YVsWmGwP .edge-depth-2{stroke-width:8;}#mermaid-svg-FHjTkKN7YVsWmGwP .section-2 line{stroke:hsl(90, 100%, 86.2745098039%);stroke-width:3;}#mermaid-svg-FHjTkKN7YVsWmGwP .disabled,#mermaid-svg-FHjTkKN7YVsWmGwP .disabled circle,#mermaid-svg-FHjTkKN7YVsWmGwP .disabled text{fill:lightgray;}#mermaid-svg-FHjTkKN7YVsWmGwP .disabled text{fill:#efefef;}#mermaid-svg-FHjTkKN7YVsWmGwP .section-3 rect,#mermaid-svg-FHjTkKN7YVsWmGwP .section-3 path,#mermaid-svg-FHjTkKN7YVsWmGwP .section-3 circle,#mermaid-svg-FHjTkKN7YVsWmGwP .section-3 polygon,#mermaid-svg-FHjTkKN7YVsWmGwP .section-3 path{fill:hsl(300, 100%, 76.2745098039%);}#mermaid-svg-FHjTkKN7YVsWmGwP .section-3 text{fill:black;}#mermaid-svg-FHjTkKN7YVsWmGwP .node-icon-3{font-size:40px;color:black;}#mermaid-svg-FHjTkKN7YVsWmGwP .section-edge-3{stroke:hsl(300, 100%, 76.2745098039%);}#mermaid-svg-FHjTkKN7YVsWmGwP .edge-depth-3{stroke-width:5;}#mermaid-svg-FHjTkKN7YVsWmGwP .section-3 line{stroke:hsl(120, 100%, 86.2745098039%);stroke-width:3;}#mermaid-svg-FHjTkKN7YVsWmGwP .disabled,#mermaid-svg-FHjTkKN7YVsWmGwP .disabled circle,#mermaid-svg-FHjTkKN7YVsWmGwP .disabled text{fill:lightgray;}#mermaid-svg-FHjTkKN7YVsWmGwP .disabled text{fill:#efefef;}#mermaid-svg-FHjTkKN7YVsWmGwP .section-4 rect,#mermaid-svg-FHjTkKN7YVsWmGwP .section-4 path,#mermaid-svg-FHjTkKN7YVsWmGwP .section-4 circle,#mermaid-svg-FHjTkKN7YVsWmGwP .section-4 polygon,#mermaid-svg-FHjTkKN7YVsWmGwP .section-4 path{fill:hsl(330, 100%, 76.2745098039%);}#mermaid-svg-FHjTkKN7YVsWmGwP .section-4 text{fill:black;}#mermaid-svg-FHjTkKN7YVsWmGwP .node-icon-4{font-size:40px;color:black;}#mermaid-svg-FHjTkKN7YVsWmGwP .section-edge-4{stroke:hsl(330, 100%, 76.2745098039%);}#mermaid-svg-FHjTkKN7YVsWmGwP .edge-depth-4{stroke-width:2;}#mermaid-svg-FHjTkKN7YVsWmGwP .section-4 line{stroke:hsl(150, 100%, 86.2745098039%);stroke-width:3;}#mermaid-svg-FHjTkKN7YVsWmGwP .disabled,#mermaid-svg-FHjTkKN7YVsWmGwP .disabled circle,#mermaid-svg-FHjTkKN7YVsWmGwP .disabled text{fill:lightgray;}#mermaid-svg-FHjTkKN7YVsWmGwP .disabled text{fill:#efefef;}#mermaid-svg-FHjTkKN7YVsWmGwP .section-5 rect,#mermaid-svg-FHjTkKN7YVsWmGwP .section-5 path,#mermaid-svg-FHjTkKN7YVsWmGwP .section-5 circle,#mermaid-svg-FHjTkKN7YVsWmGwP .section-5 polygon,#mermaid-svg-FHjTkKN7YVsWmGwP .section-5 path{fill:hsl(0, 100%, 76.2745098039%);}#mermaid-svg-FHjTkKN7YVsWmGwP .section-5 text{fill:black;}#mermaid-svg-FHjTkKN7YVsWmGwP .node-icon-5{font-size:40px;color:black;}#mermaid-svg-FHjTkKN7YVsWmGwP .section-edge-5{stroke:hsl(0, 100%, 76.2745098039%);}#mermaid-svg-FHjTkKN7YVsWmGwP .edge-depth-5{stroke-width:-1;}#mermaid-svg-FHjTkKN7YVsWmGwP .section-5 line{stroke:hsl(180, 100%, 86.2745098039%);stroke-width:3;}#mermaid-svg-FHjTkKN7YVsWmGwP .disabled,#mermaid-svg-FHjTkKN7YVsWmGwP .disabled circle,#mermaid-svg-FHjTkKN7YVsWmGwP .disabled text{fill:lightgray;}#mermaid-svg-FHjTkKN7YVsWmGwP .disabled text{fill:#efefef;}#mermaid-svg-FHjTkKN7YVsWmGwP .section-6 rect,#mermaid-svg-FHjTkKN7YVsWmGwP .section-6 path,#mermaid-svg-FHjTkKN7YVsWmGwP .section-6 circle,#mermaid-svg-FHjTkKN7YVsWmGwP .section-6 polygon,#mermaid-svg-FHjTkKN7YVsWmGwP .section-6 path{fill:hsl(30, 100%, 76.2745098039%);}#mermaid-svg-FHjTkKN7YVsWmGwP .section-6 text{fill:black;}#mermaid-svg-FHjTkKN7YVsWmGwP .node-icon-6{font-size:40px;color:black;}#mermaid-svg-FHjTkKN7YVsWmGwP .section-edge-6{stroke:hsl(30, 100%, 76.2745098039%);}#mermaid-svg-FHjTkKN7YVsWmGwP .edge-depth-6{stroke-width:-4;}#mermaid-svg-FHjTkKN7YVsWmGwP .section-6 line{stroke:hsl(210, 100%, 86.2745098039%);stroke-width:3;}#mermaid-svg-FHjTkKN7YVsWmGwP .disabled,#mermaid-svg-FHjTkKN7YVsWmGwP .disabled circle,#mermaid-svg-FHjTkKN7YVsWmGwP .disabled text{fill:lightgray;}#mermaid-svg-FHjTkKN7YVsWmGwP .disabled text{fill:#efefef;}#mermaid-svg-FHjTkKN7YVsWmGwP .section-7 rect,#mermaid-svg-FHjTkKN7YVsWmGwP .section-7 path,#mermaid-svg-FHjTkKN7YVsWmGwP .section-7 circle,#mermaid-svg-FHjTkKN7YVsWmGwP .section-7 polygon,#mermaid-svg-FHjTkKN7YVsWmGwP .section-7 path{fill:hsl(90, 100%, 76.2745098039%);}#mermaid-svg-FHjTkKN7YVsWmGwP .section-7 text{fill:black;}#mermaid-svg-FHjTkKN7YVsWmGwP .node-icon-7{font-size:40px;color:black;}#mermaid-svg-FHjTkKN7YVsWmGwP .section-edge-7{stroke:hsl(90, 100%, 76.2745098039%);}#mermaid-svg-FHjTkKN7YVsWmGwP .edge-depth-7{stroke-width:-7;}#mermaid-svg-FHjTkKN7YVsWmGwP .section-7 line{stroke:hsl(270, 100%, 86.2745098039%);stroke-width:3;}#mermaid-svg-FHjTkKN7YVsWmGwP .disabled,#mermaid-svg-FHjTkKN7YVsWmGwP .disabled circle,#mermaid-svg-FHjTkKN7YVsWmGwP .disabled text{fill:lightgray;}#mermaid-svg-FHjTkKN7YVsWmGwP .disabled text{fill:#efefef;}#mermaid-svg-FHjTkKN7YVsWmGwP .section-8 rect,#mermaid-svg-FHjTkKN7YVsWmGwP .section-8 path,#mermaid-svg-FHjTkKN7YVsWmGwP .section-8 circle,#mermaid-svg-FHjTkKN7YVsWmGwP .section-8 polygon,#mermaid-svg-FHjTkKN7YVsWmGwP .section-8 path{fill:hsl(150, 100%, 76.2745098039%);}#mermaid-svg-FHjTkKN7YVsWmGwP .section-8 text{fill:black;}#mermaid-svg-FHjTkKN7YVsWmGwP .node-icon-8{font-size:40px;color:black;}#mermaid-svg-FHjTkKN7YVsWmGwP .section-edge-8{stroke:hsl(150, 100%, 76.2745098039%);}#mermaid-svg-FHjTkKN7YVsWmGwP .edge-depth-8{stroke-width:-10;}#mermaid-svg-FHjTkKN7YVsWmGwP .section-8 line{stroke:hsl(330, 100%, 86.2745098039%);stroke-width:3;}#mermaid-svg-FHjTkKN7YVsWmGwP .disabled,#mermaid-svg-FHjTkKN7YVsWmGwP .disabled circle,#mermaid-svg-FHjTkKN7YVsWmGwP .disabled text{fill:lightgray;}#mermaid-svg-FHjTkKN7YVsWmGwP .disabled text{fill:#efefef;}#mermaid-svg-FHjTkKN7YVsWmGwP .section-9 rect,#mermaid-svg-FHjTkKN7YVsWmGwP .section-9 path,#mermaid-svg-FHjTkKN7YVsWmGwP .section-9 circle,#mermaid-svg-FHjTkKN7YVsWmGwP .section-9 polygon,#mermaid-svg-FHjTkKN7YVsWmGwP .section-9 path{fill:hsl(180, 100%, 76.2745098039%);}#mermaid-svg-FHjTkKN7YVsWmGwP .section-9 text{fill:black;}#mermaid-svg-FHjTkKN7YVsWmGwP .node-icon-9{font-size:40px;color:black;}#mermaid-svg-FHjTkKN7YVsWmGwP .section-edge-9{stroke:hsl(180, 100%, 76.2745098039%);}#mermaid-svg-FHjTkKN7YVsWmGwP .edge-depth-9{stroke-width:-13;}#mermaid-svg-FHjTkKN7YVsWmGwP .section-9 line{stroke:hsl(0, 100%, 86.2745098039%);stroke-width:3;}#mermaid-svg-FHjTkKN7YVsWmGwP .disabled,#mermaid-svg-FHjTkKN7YVsWmGwP .disabled circle,#mermaid-svg-FHjTkKN7YVsWmGwP .disabled text{fill:lightgray;}#mermaid-svg-FHjTkKN7YVsWmGwP .disabled text{fill:#efefef;}#mermaid-svg-FHjTkKN7YVsWmGwP .section-10 rect,#mermaid-svg-FHjTkKN7YVsWmGwP .section-10 path,#mermaid-svg-FHjTkKN7YVsWmGwP .section-10 circle,#mermaid-svg-FHjTkKN7YVsWmGwP .section-10 polygon,#mermaid-svg-FHjTkKN7YVsWmGwP .section-10 path{fill:hsl(210, 100%, 76.2745098039%);}#mermaid-svg-FHjTkKN7YVsWmGwP .section-10 text{fill:black;}#mermaid-svg-FHjTkKN7YVsWmGwP .node-icon-10{font-size:40px;color:black;}#mermaid-svg-FHjTkKN7YVsWmGwP .section-edge-10{stroke:hsl(210, 100%, 76.2745098039%);}#mermaid-svg-FHjTkKN7YVsWmGwP .edge-depth-10{stroke-width:-16;}#mermaid-svg-FHjTkKN7YVsWmGwP .section-10 line{stroke:hsl(30, 100%, 86.2745098039%);stroke-width:3;}#mermaid-svg-FHjTkKN7YVsWmGwP .disabled,#mermaid-svg-FHjTkKN7YVsWmGwP .disabled circle,#mermaid-svg-FHjTkKN7YVsWmGwP .disabled text{fill:lightgray;}#mermaid-svg-FHjTkKN7YVsWmGwP .disabled text{fill:#efefef;}#mermaid-svg-FHjTkKN7YVsWmGwP .section-root rect,#mermaid-svg-FHjTkKN7YVsWmGwP .section-root path,#mermaid-svg-FHjTkKN7YVsWmGwP .section-root circle,#mermaid-svg-FHjTkKN7YVsWmGwP .section-root polygon{fill:hsl(240, 100%, 46.2745098039%);}#mermaid-svg-FHjTkKN7YVsWmGwP .section-root text{fill:#ffffff;}#mermaid-svg-FHjTkKN7YVsWmGwP .section-root span{color:#ffffff;}#mermaid-svg-FHjTkKN7YVsWmGwP .section-2 span{color:#ffffff;}#mermaid-svg-FHjTkKN7YVsWmGwP .icon-container{height:100%;display:flex;justify-content:center;align-items:center;}#mermaid-svg-FHjTkKN7YVsWmGwP .edge{fill:none;}#mermaid-svg-FHjTkKN7YVsWmGwP .mindmap-node-label{dy:1em;alignment-baseline:middle;text-anchor:middle;dominant-baseline:middle;text-align:center;}#mermaid-svg-FHjTkKN7YVsWmGwP :root{--mermaid-font-family:"trebuchet ms",verdana,arial,sans-serif;} K-MARVEL耐药基因检测研究
研究背景
抗微生物耐药全球威胁
短读组装耗时且长读错误高
原始读长直接检测需求
方法框架
蛋白空间k-mer匹配扩展
四模块分箱与变异过滤
原生与BLAST双重注释
核心结果
短读长F1达0.976
长读长F1达0.958
超越组装与其它分类器
应用价值
临床宏基因组监测
便携设备实时溯源
多变异与突变基因重建

背景介绍

AMR已成为全球公共卫生核心威胁，其传播依赖于抗生素选择压力下ARG在微生物种群内的扩增与水平转移。二代短读长测序受100至300碱基读长限制，全基因组重建常呈碎片化；三代长读长如ONT可覆盖完整基因与操纵子，但插入缺失错误尤其是同聚物区域的误差会掩盖染色体内耐药点突变，例如gyrA、rpoB等靶基因变异。现有计算路线主要分为两类：其一是先组装后注释，以Resistance Gene Identifier（RGI）对接CARD数据库为代表，精度较高但计算开销大；其二是读长直接比对或核苷酸k-mer分类，如SRST2、KmerResistance、ABRICATE等，但在长读错误率与未知样本中泛化有限。PointFinder等点突变工具需预先指定菌种，限制了复杂或培养阴性样本的应用。针对上述短板，作者在蛋白水平构建k-mer匹配与扩展策略，开发无组装、跨平台、可同时报告同源ARG与结构基因突变的K-MARVEL系统，并在大规模经验数据集上完成精度与资源消耗基准。

重要结果

结果1 网格搜索优化短读与长读k-mer及扩展参数

作者选取12个同时具备短读、长读与参考水平组装的菌种样本，以CARD同源模型蛋白为金标准，并对变异与过表达模型采用RGI宽松模式调用，设定一致性与覆盖度阈值均为90%。对k-mer长度6至10、扩展长度6至12进行全网格组合，按12个样本聚合真阳性、假阳性与假阴性并计算F1。长读长数据最优组合为k-mer 7、扩展12，短读长数据最优组合为k-mer 6、扩展9；该参数下长读全域F1为0.958、其中变异与过表达模型子集F1为0.962，短读全域F1为0.976、变异与过表达子集F1为0.982，后续经验基准均沿用上述最优参数。

结果2 经验数据集全局分类性能与同源及突变模型检测

作者将评估扩展至209个长读与205个短读数据集，覆盖22个系统发育距离较远的菌种，多数属于世界卫生组织耐药优先病原体清单。所有样本均以前参考组装调用ARG作为金标准。长读整体F1为0.958、精确率0.948至0.959、召回率0.968至0.965；短读整体F1为0.976、精确率0.969、召回率0.983至0.996。对变异与过表达模型，长读F1为0.962、短读F1为0.982，表明蛋白k-mer扩展可重建含多个耐药错义突变的持家基因与调控基因，而不仅限于传统同源ARG。

图1图例：A为短读数据在22种细菌数据集上各ARG分类器准确性箱线图与折线，样本量205；水平虚线表示0.9准确性阈值，箱线中心为中位数，箱体表示第25与75四分位数，须线覆盖1.5倍四分位距内极值，离群点单独绘制；K-MARVEL在21/22物种准确性≥90%。B为长读数据在209个数据集上各长读分类器准确性箱线图与折线，K-MARVEL（NATIVE与BLAST）中位数高于ABRICATE与KARGA，在13/22物种准确性≥90%；整体显示蛋白空间分类器在不同菌种与平台均保持稳健。

结果3 与主流短读及长读ARG分类器的精度基准对比

在以CARD同源模型为对象的横向基准中，短读对照包括ARGprofiler、DeepARG、KARGA、KmerResistance、RGI-BWT、ShortBRED与SRST2，长读对照包括ABRICATE与KARGA。K-MARVEL（BLAST）短读平均F1为0.979、精确率0.971、召回率0.986；K-MARVEL（NATIVE）短读平均F1为0.974。长读K-MARVEL（BLAST）平均F1为0.961、NATIVE为0.952。全局汇总显示RGI-BWT短读F1为0.895、SRST2为0.853、ABRICATE长读F1为0.521、KARGA长读F1为0.333；K-MARVEL均显著占优。假阳性分析进一步显示，部分被RGI忽略的多变异ARG、低一致度但可被k-mer重建的片段基因，以及组装断裂基因，均可由K-MARVEL召回并写入最终输出。

结果4 计算资源消耗基准：运行时间与内存占用

作者在16线程AMD EPYC 7543、300GB内存服务器上比较读长级与组装级流程。短读读长级工具中RGI-BWT中位运行11.68秒、ARGprofiler 13.10秒，K-MARVEL中位26.92秒；K-MARVEL中位峰值内存0.82GB，高于ARGprofiler的0.11GB但与多数深度学习工具相比更可控。长读中K-MARVEL中位运行50.57秒，快于ABRICATE与KARGA；中位峰值内存2.07GB，低于常规组装后注释但高于ABRICATE的0.74GB。与Flye或SPAdes组装后再用RGI注释相比，短读组装流程中位290.44秒、K-MARVEL仅26.92秒；长读组装流程中位684.46秒、K-MARVEL仅50.57秒；内存方面短读组装中位6.471GB、K-MARVEL 0.82GB，长读组装内存显著更高而K-MARVEL中位约2.11GB；Wilcoxon秩和检验显示差异均极显著。

图2图例：A短读各分类器执行时间箱线图，含RGI-BWT、ARGprofiler、K-MARVEL等，K-MARVEL中位26.92秒；B短读各分类器峰值内存箱线图，K-MARVEL中位0.82GB；C长读各分类器执行时间箱线图，K-MARVEL中位50.57秒且快于ABRICATE与KARGA，配对调整p值ABRICATE为1.31×10⁻¹⁷、KARGA为2.04×10⁻³¹；D长读各分类器峰值内存箱线图，ABRICATE中位0.74GB、K-MARVEL中位2.07GB；E组装流程与K-MARVEL执行时间对比，短读调整p值2.38×10⁻⁶⁶、长读2.42×10⁻⁶⁰；F组装流程与K-MARVEL峰值内存对比，短读调整p值4.58×10⁻⁶³、长读4.72×10⁻¹⁷；所有测试基于16线程服务器。

结果5 大规模数据集可扩展性与资源线性评估

短读数据集规模0.79至34.26百万读长，K-MARVEL运行6.7至208.49秒、峰值内存0.54至9.1GB；长读数据集规模4174至261万读长，运行4.78至1571.14秒、峰值内存0.89至62.117GB。运行时间随读长数量呈线性增长，短读Pearson R=0.87、长读R=0.64；随总碱基数亦近线性，短读R=0.88、长读R=0.85。内存与数据规模无显著相关，短读R=-0.075、长读R=-0.034，回归线接近平坦，说明扩大样本量不会成比例增加内存压力。该特征使K-MARVEL适用于便携式笔记本与床旁测序环境。

图3图例：A执行时间随数据集规模变化散点与直线拟合，长读调整p值2.42×10⁻⁶⁰、短读调整p值2.38×10⁻⁶⁶，显示短读与长读均近线性扩展；B峰值内存随数据集规模变化散点与直线拟合，长读调整p值0.625、短读调整p值0.566，回归接近平坦，表明短读与长读内存消耗整体低位稳定，不随读长数或碱基数成比例上升。

结果6 四模块蛋白k-mer流程、变异重建与嵌合过滤

K-MARVEL以Rust实现，包含参考索引与六框匹配、ARG组装、假变异与持家基因过滤、嵌合体移除四个模块。模块一将参考ARG蛋白按k-mer构建哈希索引，读长翻译为六框蛋白k-mer并完美匹配后向两端扩展；仅保留单框内至少5个重叠扩展k-mer的读长。模块二按参考坐标组装共识蛋白，计算覆盖度与Smith-Waterman一致性，未覆盖坐标以符号占位。模块三以复合得分、最长公共子序列与已知突变集筛选假变异，并二次收录含点突变的持家或结构基因；变异模型与过表达模型由此可直接输出。模块四按同家族k-mer共享率≥95%且独特读长率＜5%剔除嵌合假基因。以痢疾杆菌sul2为例，组装法因低覆盖仅得两段碎片，K-MARVEL可借全基因k-mer重建完整基因。最终输出JSON与TSV，并对同源模型基因提供原生注释与蛋白BLAST重注释。

图4图例：K-MARVEL工作流示意图。模块一展示参考ARG蛋白库索引、原始FASTQ六框翻译、蛋白k-mer完美匹配与双向扩展；模块二展示按参考位置保留最佳读长框、组装共识序列、计算覆盖度与一致性并过滤低质基因；模块三展示复合得分过滤假变异、成对读长重叠去冗余、以及针对含耐药突变持家基因与过表达调控基因的第二轮筛查；模块四展示同家族独特k-mer比例与独特读长比例判定并移除嵌合体；末端输出JSON与TSV，并可调用蛋白BLAST生成补充注释。示意图使用Figma制作。

方法学参考

#mermaid-svg-WS5w1K0qACIm0cHV{font-family:"trebuchet ms",verdana,arial,sans-serif;font-size:16px;fill:#333;}@keyframes edge-animation-frame{from{stroke-dashoffset:0;}}@keyframes dash{to{stroke-dashoffset:0;}}#mermaid-svg-WS5w1K0qACIm0cHV .edge-animation-slow{stroke-dasharray:9,5!important;stroke-dashoffset:900;animation:dash 50s linear infinite;stroke-linecap:round;}#mermaid-svg-WS5w1K0qACIm0cHV .edge-animation-fast{stroke-dasharray:9,5!important;stroke-dashoffset:900;animation:dash 20s linear infinite;stroke-linecap:round;}#mermaid-svg-WS5w1K0qACIm0cHV .error-icon{fill:#552222;}#mermaid-svg-WS5w1K0qACIm0cHV .error-text{fill:#552222;stroke:#552222;}#mermaid-svg-WS5w1K0qACIm0cHV .edge-thickness-normal{stroke-width:1px;}#mermaid-svg-WS5w1K0qACIm0cHV .edge-thickness-thick{stroke-width:3.5px;}#mermaid-svg-WS5w1K0qACIm0cHV .edge-pattern-solid{stroke-dasharray:0;}#mermaid-svg-WS5w1K0qACIm0cHV .edge-thickness-invisible{stroke-width:0;fill:none;}#mermaid-svg-WS5w1K0qACIm0cHV .edge-pattern-dashed{stroke-dasharray:3;}#mermaid-svg-WS5w1K0qACIm0cHV .edge-pattern-dotted{stroke-dasharray:2;}#mermaid-svg-WS5w1K0qACIm0cHV .marker{fill:#333333;stroke:#333333;}#mermaid-svg-WS5w1K0qACIm0cHV .marker.cross{stroke:#333333;}#mermaid-svg-WS5w1K0qACIm0cHV svg{font-family:"trebuchet ms",verdana,arial,sans-serif;font-size:16px;}#mermaid-svg-WS5w1K0qACIm0cHV p{margin:0;}#mermaid-svg-WS5w1K0qACIm0cHV .label{font-family:"trebuchet ms",verdana,arial,sans-serif;color:#333;}#mermaid-svg-WS5w1K0qACIm0cHV .cluster-label text{fill:#333;}#mermaid-svg-WS5w1K0qACIm0cHV .cluster-label span{color:#333;}#mermaid-svg-WS5w1K0qACIm0cHV .cluster-label span p{background-color:transparent;}#mermaid-svg-WS5w1K0qACIm0cHV .label text,#mermaid-svg-WS5w1K0qACIm0cHV span{fill:#333;color:#333;}#mermaid-svg-WS5w1K0qACIm0cHV .node rect,#mermaid-svg-WS5w1K0qACIm0cHV .node circle,#mermaid-svg-WS5w1K0qACIm0cHV .node ellipse,#mermaid-svg-WS5w1K0qACIm0cHV .node polygon,#mermaid-svg-WS5w1K0qACIm0cHV .node path{fill:#ECECFF;stroke:#9370DB;stroke-width:1px;}#mermaid-svg-WS5w1K0qACIm0cHV .rough-node .label text,#mermaid-svg-WS5w1K0qACIm0cHV .node .label text,#mermaid-svg-WS5w1K0qACIm0cHV .image-shape .label,#mermaid-svg-WS5w1K0qACIm0cHV .icon-shape .label{text-anchor:middle;}#mermaid-svg-WS5w1K0qACIm0cHV .node .katex path{fill:#000;stroke:#000;stroke-width:1px;}#mermaid-svg-WS5w1K0qACIm0cHV .rough-node .label,#mermaid-svg-WS5w1K0qACIm0cHV .node .label,#mermaid-svg-WS5w1K0qACIm0cHV .image-shape .label,#mermaid-svg-WS5w1K0qACIm0cHV .icon-shape .label{text-align:center;}#mermaid-svg-WS5w1K0qACIm0cHV .node.clickable{cursor:pointer;}#mermaid-svg-WS5w1K0qACIm0cHV .root .anchor path{fill:#333333!important;stroke-width:0;stroke:#333333;}#mermaid-svg-WS5w1K0qACIm0cHV .arrowheadPath{fill:#333333;}#mermaid-svg-WS5w1K0qACIm0cHV .edgePath .path{stroke:#333333;stroke-width:2.0px;}#mermaid-svg-WS5w1K0qACIm0cHV .flowchart-link{stroke:#333333;fill:none;}#mermaid-svg-WS5w1K0qACIm0cHV .edgeLabel{background-color:rgba(232,232,232, 0.8);text-align:center;}#mermaid-svg-WS5w1K0qACIm0cHV .edgeLabel p{background-color:rgba(232,232,232, 0.8);}#mermaid-svg-WS5w1K0qACIm0cHV .edgeLabel rect{opacity:0.5;background-color:rgba(232,232,232, 0.8);fill:rgba(232,232,232, 0.8);}#mermaid-svg-WS5w1K0qACIm0cHV .labelBkg{background-color:rgba(232, 232, 232, 0.5);}#mermaid-svg-WS5w1K0qACIm0cHV .cluster rect{fill:#ffffde;stroke:#aaaa33;stroke-width:1px;}#mermaid-svg-WS5w1K0qACIm0cHV .cluster text{fill:#333;}#mermaid-svg-WS5w1K0qACIm0cHV .cluster span{color:#333;}#mermaid-svg-WS5w1K0qACIm0cHV div.mermaidTooltip{position:absolute;text-align:center;max-width:200px;padding:2px;font-family:"trebuchet ms",verdana,arial,sans-serif;font-size:12px;background:hsl(80, 100%, 96.2745098039%);border:1px solid #aaaa33;border-radius:2px;pointer-events:none;z-index:100;}#mermaid-svg-WS5w1K0qACIm0cHV .flowchartTitleText{text-anchor:middle;font-size:18px;fill:#333;}#mermaid-svg-WS5w1K0qACIm0cHV rect.text{fill:none;stroke-width:0;}#mermaid-svg-WS5w1K0qACIm0cHV .icon-shape,#mermaid-svg-WS5w1K0qACIm0cHV .image-shape{background-color:rgba(232,232,232, 0.8);text-align:center;}#mermaid-svg-WS5w1K0qACIm0cHV .icon-shape p,#mermaid-svg-WS5w1K0qACIm0cHV .image-shape p{background-color:rgba(232,232,232, 0.8);padding:2px;}#mermaid-svg-WS5w1K0qACIm0cHV .icon-shape .label rect,#mermaid-svg-WS5w1K0qACIm0cHV .image-shape .label rect{opacity:0.5;background-color:rgba(232,232,232, 0.8);fill:rgba(232,232,232, 0.8);}#mermaid-svg-WS5w1K0qACIm0cHV .label-icon{display:inline-block;height:1em;overflow:visible;vertical-align:-0.125em;}#mermaid-svg-WS5w1K0qACIm0cHV .node .label-icon path{fill:currentColor;stroke:revert;stroke-width:revert;}#mermaid-svg-WS5w1K0qACIm0cHV :root{--mermaid-font-family:"trebuchet ms",verdana,arial,sans-serif;} 原始FASTQ短读或长读输入
参考ARG蛋白库构建蛋白k-mer哈希索引
读长六框翻译并匹配k-mer向两端扩展
筛选每框至少5个重叠扩展k-mer的最佳框
按参考坐标组装共识蛋白并计算覆盖度与一致性
复合得分过滤假变异并筛查突变型持家基因
同家族k-mer与读长重叠分析移除嵌合体
可选蛋白BLAST重注释并输出JSON与TSV

上述流程中，参考索引默认采用CARD蛋白序列，也可替换为用户维护的ARG库；六框翻译可吸收核苷酸替换错误，扩展k-mer可跨越低覆盖断裂区域。复合得分由一致性、k-mer比率与归一化比对得分三等分平均构成，配合90%一致度与覆盖度阈值可降低长读插入缺失造成的假变异。该框架可迁移至环境宏基因组、临床尿液或血液游离DNA中的耐药基因快速普查。

总结

作者开发的K-MARVEL以蛋白k-mer匹配扩展替代传统组装与核苷酸比对，在22个菌种、超过400个经验数据集中实现了短读F1 0.976与长读F1 0.958的同源ARG检测精度，并额外覆盖变异、过表达及多变异ARG。与RGI-BWT、SRST2、ABRICATE等工具相比，其在保证精度的同时显著降低运行时间与内存占用，且对片段化基因具有重建能力。局限性在于其高度依赖参考数据库，难以发现全新耐药决定子，且当前不直接输出ARG所属病原种系，需配合分类或分箱流程。未来在复杂临床宏基因组、 duplex纳米孔长读与便携测序设备中的集成，可进一步提升低覆盖、高宿主背景样本的抗耐药监测效率。

参考文献

Mahar N. S., Branders S., Grabherr M. G., Gupta I., Ahmad R. (2026). K-MARVEL: K-Mer-based antimicrobial resistance virtual exploration lab. Nature Communications. https://doi.org/10.1038/s41467-026-77438-8
K-MARVEL开源代码库：https://bitbucket.org/amr-avengers/k-marvel/src/main/
Comprehensive Antibiotic Resistance Database（CARD）：https://card.mcmaster.ca/
NCBI Sequence Read Archive（SRA）数据集索引见补充数据10：https://www.ncbi.nlm.nih.gov/sra
Resistance Gene Identifier（RGI）与CARD注释流程：https://github.com/arpcard/rgi
PointFinder染色体点突变工具：https://github.com/pointfinder-org/PointFinder
ABRICATE抗性基因筛查：https://github.com/tseemann/abricate
牛津纳米孔测序与duplex basecalling技术说明：https://nanoporetech.com/