Microbiome｜微生物组中介分析的错误控制：系统基准测试与 CAMRA 补救框架

摘要

背景：识别能够介导"暴露（exposure）对健康结局（outcome）效应"的微生物类群，有助于阐明因果通路并提示治疗靶点。微生物组测序通常测得类群的相对丰度（relative abundance，RA） ，而机制解释往往更关心并未被直接观测到的绝对丰度（absolute abundance，AA）。虽然组成性（compositionality）导致的假阳性已在差异丰度等分析中受到关注，但微生物组中介方法的系统基准测试仍十分有限。

结果：作者基于大规模模拟与真实数据分析，对九种中介分析方法进行基准测试；模拟以实验恢复的 AA 模板为锚点，使生成的 AA 轮廓比常用模拟器更好地保留经验结构。多种方法在模拟中表现出错误率膨胀，且组成效应越强，膨胀越严重。在真实数据中，模拟里严重膨胀的方法也返回了明显更大的中介类群集合。作为补救，作者提出 CAMRA（Causal Absolute-abundance Mediation from Relative-Abundance data），可从标准 RA 数据推断并检验 AA 水平的中介效应，在分类单元水平中介发现中改善 FDR 校准与统计功效，同时保持较好的运行效率。

结论：可靠的微生物中介发现依赖于校准良好的统计推断。该基准揭示了 RA 水平中介检验的系统性挑战，而 CAMRA 通过面向 AA 水平中介效应提供了补救方案。

keywords：绝对丰度；组成效应；错误发现；微生物组中介检验；相对丰度

文献信息

Wang, Q., Li, Y., Peng, Y., & Tang, Z.-Z. (2026). Error control for microbiome mediation analysis: benchmarking and remedy. Microbiome. https://doi.org/10.1186/s40168-026-02527-1
期刊：Microbiome（IF=14.9）
接收时间：2026 年 8 月 11 日；在线发表：2026 年 9 月 25 日（Article in Press，未经最终润色版）

研究总结

#mermaid-svg-m50KfoYxo9Y4mDvI{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-m50KfoYxo9Y4mDvI .edge-animation-slow{stroke-dasharray:9,5!important;stroke-dashoffset:900;animation:dash 50s linear infinite;stroke-linecap:round;}#mermaid-svg-m50KfoYxo9Y4mDvI .edge-animation-fast{stroke-dasharray:9,5!important;stroke-dashoffset:900;animation:dash 20s linear infinite;stroke-linecap:round;}#mermaid-svg-m50KfoYxo9Y4mDvI .error-icon{fill:#552222;}#mermaid-svg-m50KfoYxo9Y4mDvI .error-text{fill:#552222;stroke:#552222;}#mermaid-svg-m50KfoYxo9Y4mDvI .edge-thickness-normal{stroke-width:1px;}#mermaid-svg-m50KfoYxo9Y4mDvI .edge-thickness-thick{stroke-width:3.5px;}#mermaid-svg-m50KfoYxo9Y4mDvI .edge-pattern-solid{stroke-dasharray:0;}#mermaid-svg-m50KfoYxo9Y4mDvI .edge-thickness-invisible{stroke-width:0;fill:none;}#mermaid-svg-m50KfoYxo9Y4mDvI .edge-pattern-dashed{stroke-dasharray:3;}#mermaid-svg-m50KfoYxo9Y4mDvI .edge-pattern-dotted{stroke-dasharray:2;}#mermaid-svg-m50KfoYxo9Y4mDvI .marker{fill:#333333;stroke:#333333;}#mermaid-svg-m50KfoYxo9Y4mDvI .marker.cross{stroke:#333333;}#mermaid-svg-m50KfoYxo9Y4mDvI svg{font-family:"trebuchet ms",verdana,arial,sans-serif;font-size:16px;}#mermaid-svg-m50KfoYxo9Y4mDvI p{margin:0;}#mermaid-svg-m50KfoYxo9Y4mDvI .edge{stroke-width:3;}#mermaid-svg-m50KfoYxo9Y4mDvI .section--1 rect,#mermaid-svg-m50KfoYxo9Y4mDvI .section--1 path,#mermaid-svg-m50KfoYxo9Y4mDvI .section--1 circle,#mermaid-svg-m50KfoYxo9Y4mDvI .section--1 polygon,#mermaid-svg-m50KfoYxo9Y4mDvI .section--1 path{fill:hsl(240, 100%, 76.2745098039%);}#mermaid-svg-m50KfoYxo9Y4mDvI .section--1 text{fill:#ffffff;}#mermaid-svg-m50KfoYxo9Y4mDvI .node-icon--1{font-size:40px;color:#ffffff;}#mermaid-svg-m50KfoYxo9Y4mDvI .section-edge--1{stroke:hsl(240, 100%, 76.2745098039%);}#mermaid-svg-m50KfoYxo9Y4mDvI .edge-depth--1{stroke-width:17;}#mermaid-svg-m50KfoYxo9Y4mDvI .section--1 line{stroke:hsl(60, 100%, 86.2745098039%);stroke-width:3;}#mermaid-svg-m50KfoYxo9Y4mDvI .disabled,#mermaid-svg-m50KfoYxo9Y4mDvI .disabled circle,#mermaid-svg-m50KfoYxo9Y4mDvI .disabled text{fill:lightgray;}#mermaid-svg-m50KfoYxo9Y4mDvI .disabled text{fill:#efefef;}#mermaid-svg-m50KfoYxo9Y4mDvI .section-0 rect,#mermaid-svg-m50KfoYxo9Y4mDvI .section-0 path,#mermaid-svg-m50KfoYxo9Y4mDvI .section-0 circle,#mermaid-svg-m50KfoYxo9Y4mDvI .section-0 polygon,#mermaid-svg-m50KfoYxo9Y4mDvI .section-0 path{fill:hsl(60, 100%, 73.5294117647%);}#mermaid-svg-m50KfoYxo9Y4mDvI .section-0 text{fill:black;}#mermaid-svg-m50KfoYxo9Y4mDvI .node-icon-0{font-size:40px;color:black;}#mermaid-svg-m50KfoYxo9Y4mDvI .section-edge-0{stroke:hsl(60, 100%, 73.5294117647%);}#mermaid-svg-m50KfoYxo9Y4mDvI .edge-depth-0{stroke-width:14;}#mermaid-svg-m50KfoYxo9Y4mDvI .section-0 line{stroke:hsl(240, 100%, 83.5294117647%);stroke-width:3;}#mermaid-svg-m50KfoYxo9Y4mDvI .disabled,#mermaid-svg-m50KfoYxo9Y4mDvI .disabled circle,#mermaid-svg-m50KfoYxo9Y4mDvI .disabled text{fill:lightgray;}#mermaid-svg-m50KfoYxo9Y4mDvI .disabled text{fill:#efefef;}#mermaid-svg-m50KfoYxo9Y4mDvI .section-1 rect,#mermaid-svg-m50KfoYxo9Y4mDvI .section-1 path,#mermaid-svg-m50KfoYxo9Y4mDvI .section-1 circle,#mermaid-svg-m50KfoYxo9Y4mDvI .section-1 polygon,#mermaid-svg-m50KfoYxo9Y4mDvI .section-1 path{fill:hsl(80, 100%, 76.2745098039%);}#mermaid-svg-m50KfoYxo9Y4mDvI .section-1 text{fill:black;}#mermaid-svg-m50KfoYxo9Y4mDvI .node-icon-1{font-size:40px;color:black;}#mermaid-svg-m50KfoYxo9Y4mDvI .section-edge-1{stroke:hsl(80, 100%, 76.2745098039%);}#mermaid-svg-m50KfoYxo9Y4mDvI .edge-depth-1{stroke-width:11;}#mermaid-svg-m50KfoYxo9Y4mDvI .section-1 line{stroke:hsl(260, 100%, 86.2745098039%);stroke-width:3;}#mermaid-svg-m50KfoYxo9Y4mDvI .disabled,#mermaid-svg-m50KfoYxo9Y4mDvI .disabled circle,#mermaid-svg-m50KfoYxo9Y4mDvI .disabled text{fill:lightgray;}#mermaid-svg-m50KfoYxo9Y4mDvI .disabled text{fill:#efefef;}#mermaid-svg-m50KfoYxo9Y4mDvI .section-2 rect,#mermaid-svg-m50KfoYxo9Y4mDvI .section-2 path,#mermaid-svg-m50KfoYxo9Y4mDvI .section-2 circle,#mermaid-svg-m50KfoYxo9Y4mDvI .section-2 polygon,#mermaid-svg-m50KfoYxo9Y4mDvI .section-2 path{fill:hsl(270, 100%, 76.2745098039%);}#mermaid-svg-m50KfoYxo9Y4mDvI .section-2 text{fill:#ffffff;}#mermaid-svg-m50KfoYxo9Y4mDvI .node-icon-2{font-size:40px;color:#ffffff;}#mermaid-svg-m50KfoYxo9Y4mDvI .section-edge-2{stroke:hsl(270, 100%, 76.2745098039%);}#mermaid-svg-m50KfoYxo9Y4mDvI .edge-depth-2{stroke-width:8;}#mermaid-svg-m50KfoYxo9Y4mDvI .section-2 line{stroke:hsl(90, 100%, 86.2745098039%);stroke-width:3;}#mermaid-svg-m50KfoYxo9Y4mDvI .disabled,#mermaid-svg-m50KfoYxo9Y4mDvI .disabled circle,#mermaid-svg-m50KfoYxo9Y4mDvI .disabled text{fill:lightgray;}#mermaid-svg-m50KfoYxo9Y4mDvI .disabled text{fill:#efefef;}#mermaid-svg-m50KfoYxo9Y4mDvI .section-3 rect,#mermaid-svg-m50KfoYxo9Y4mDvI .section-3 path,#mermaid-svg-m50KfoYxo9Y4mDvI .section-3 circle,#mermaid-svg-m50KfoYxo9Y4mDvI .section-3 polygon,#mermaid-svg-m50KfoYxo9Y4mDvI .section-3 path{fill:hsl(300, 100%, 76.2745098039%);}#mermaid-svg-m50KfoYxo9Y4mDvI .section-3 text{fill:black;}#mermaid-svg-m50KfoYxo9Y4mDvI .node-icon-3{font-size:40px;color:black;}#mermaid-svg-m50KfoYxo9Y4mDvI .section-edge-3{stroke:hsl(300, 100%, 76.2745098039%);}#mermaid-svg-m50KfoYxo9Y4mDvI .edge-depth-3{stroke-width:5;}#mermaid-svg-m50KfoYxo9Y4mDvI .section-3 line{stroke:hsl(120, 100%, 86.2745098039%);stroke-width:3;}#mermaid-svg-m50KfoYxo9Y4mDvI .disabled,#mermaid-svg-m50KfoYxo9Y4mDvI .disabled circle,#mermaid-svg-m50KfoYxo9Y4mDvI .disabled text{fill:lightgray;}#mermaid-svg-m50KfoYxo9Y4mDvI .disabled text{fill:#efefef;}#mermaid-svg-m50KfoYxo9Y4mDvI .section-4 rect,#mermaid-svg-m50KfoYxo9Y4mDvI .section-4 path,#mermaid-svg-m50KfoYxo9Y4mDvI .section-4 circle,#mermaid-svg-m50KfoYxo9Y4mDvI .section-4 polygon,#mermaid-svg-m50KfoYxo9Y4mDvI .section-4 path{fill:hsl(330, 100%, 76.2745098039%);}#mermaid-svg-m50KfoYxo9Y4mDvI .section-4 text{fill:black;}#mermaid-svg-m50KfoYxo9Y4mDvI .node-icon-4{font-size:40px;color:black;}#mermaid-svg-m50KfoYxo9Y4mDvI .section-edge-4{stroke:hsl(330, 100%, 76.2745098039%);}#mermaid-svg-m50KfoYxo9Y4mDvI .edge-depth-4{stroke-width:2;}#mermaid-svg-m50KfoYxo9Y4mDvI .section-4 line{stroke:hsl(150, 100%, 86.2745098039%);stroke-width:3;}#mermaid-svg-m50KfoYxo9Y4mDvI .disabled,#mermaid-svg-m50KfoYxo9Y4mDvI .disabled circle,#mermaid-svg-m50KfoYxo9Y4mDvI .disabled text{fill:lightgray;}#mermaid-svg-m50KfoYxo9Y4mDvI .disabled text{fill:#efefef;}#mermaid-svg-m50KfoYxo9Y4mDvI .section-5 rect,#mermaid-svg-m50KfoYxo9Y4mDvI .section-5 path,#mermaid-svg-m50KfoYxo9Y4mDvI .section-5 circle,#mermaid-svg-m50KfoYxo9Y4mDvI .section-5 polygon,#mermaid-svg-m50KfoYxo9Y4mDvI .section-5 path{fill:hsl(0, 100%, 76.2745098039%);}#mermaid-svg-m50KfoYxo9Y4mDvI .section-5 text{fill:black;}#mermaid-svg-m50KfoYxo9Y4mDvI .node-icon-5{font-size:40px;color:black;}#mermaid-svg-m50KfoYxo9Y4mDvI .section-edge-5{stroke:hsl(0, 100%, 76.2745098039%);}#mermaid-svg-m50KfoYxo9Y4mDvI .edge-depth-5{stroke-width:-1;}#mermaid-svg-m50KfoYxo9Y4mDvI .section-5 line{stroke:hsl(180, 100%, 86.2745098039%);stroke-width:3;}#mermaid-svg-m50KfoYxo9Y4mDvI .disabled,#mermaid-svg-m50KfoYxo9Y4mDvI .disabled circle,#mermaid-svg-m50KfoYxo9Y4mDvI .disabled text{fill:lightgray;}#mermaid-svg-m50KfoYxo9Y4mDvI .disabled text{fill:#efefef;}#mermaid-svg-m50KfoYxo9Y4mDvI .section-6 rect,#mermaid-svg-m50KfoYxo9Y4mDvI .section-6 path,#mermaid-svg-m50KfoYxo9Y4mDvI .section-6 circle,#mermaid-svg-m50KfoYxo9Y4mDvI .section-6 polygon,#mermaid-svg-m50KfoYxo9Y4mDvI .section-6 path{fill:hsl(30, 100%, 76.2745098039%);}#mermaid-svg-m50KfoYxo9Y4mDvI .section-6 text{fill:black;}#mermaid-svg-m50KfoYxo9Y4mDvI .node-icon-6{font-size:40px;color:black;}#mermaid-svg-m50KfoYxo9Y4mDvI .section-edge-6{stroke:hsl(30, 100%, 76.2745098039%);}#mermaid-svg-m50KfoYxo9Y4mDvI .edge-depth-6{stroke-width:-4;}#mermaid-svg-m50KfoYxo9Y4mDvI .section-6 line{stroke:hsl(210, 100%, 86.2745098039%);stroke-width:3;}#mermaid-svg-m50KfoYxo9Y4mDvI .disabled,#mermaid-svg-m50KfoYxo9Y4mDvI .disabled circle,#mermaid-svg-m50KfoYxo9Y4mDvI .disabled text{fill:lightgray;}#mermaid-svg-m50KfoYxo9Y4mDvI .disabled text{fill:#efefef;}#mermaid-svg-m50KfoYxo9Y4mDvI .section-7 rect,#mermaid-svg-m50KfoYxo9Y4mDvI .section-7 path,#mermaid-svg-m50KfoYxo9Y4mDvI .section-7 circle,#mermaid-svg-m50KfoYxo9Y4mDvI .section-7 polygon,#mermaid-svg-m50KfoYxo9Y4mDvI .section-7 path{fill:hsl(90, 100%, 76.2745098039%);}#mermaid-svg-m50KfoYxo9Y4mDvI .section-7 text{fill:black;}#mermaid-svg-m50KfoYxo9Y4mDvI .node-icon-7{font-size:40px;color:black;}#mermaid-svg-m50KfoYxo9Y4mDvI .section-edge-7{stroke:hsl(90, 100%, 76.2745098039%);}#mermaid-svg-m50KfoYxo9Y4mDvI .edge-depth-7{stroke-width:-7;}#mermaid-svg-m50KfoYxo9Y4mDvI .section-7 line{stroke:hsl(270, 100%, 86.2745098039%);stroke-width:3;}#mermaid-svg-m50KfoYxo9Y4mDvI .disabled,#mermaid-svg-m50KfoYxo9Y4mDvI .disabled circle,#mermaid-svg-m50KfoYxo9Y4mDvI .disabled text{fill:lightgray;}#mermaid-svg-m50KfoYxo9Y4mDvI .disabled text{fill:#efefef;}#mermaid-svg-m50KfoYxo9Y4mDvI .section-8 rect,#mermaid-svg-m50KfoYxo9Y4mDvI .section-8 path,#mermaid-svg-m50KfoYxo9Y4mDvI .section-8 circle,#mermaid-svg-m50KfoYxo9Y4mDvI .section-8 polygon,#mermaid-svg-m50KfoYxo9Y4mDvI .section-8 path{fill:hsl(150, 100%, 76.2745098039%);}#mermaid-svg-m50KfoYxo9Y4mDvI .section-8 text{fill:black;}#mermaid-svg-m50KfoYxo9Y4mDvI .node-icon-8{font-size:40px;color:black;}#mermaid-svg-m50KfoYxo9Y4mDvI .section-edge-8{stroke:hsl(150, 100%, 76.2745098039%);}#mermaid-svg-m50KfoYxo9Y4mDvI .edge-depth-8{stroke-width:-10;}#mermaid-svg-m50KfoYxo9Y4mDvI .section-8 line{stroke:hsl(330, 100%, 86.2745098039%);stroke-width:3;}#mermaid-svg-m50KfoYxo9Y4mDvI .disabled,#mermaid-svg-m50KfoYxo9Y4mDvI .disabled circle,#mermaid-svg-m50KfoYxo9Y4mDvI .disabled text{fill:lightgray;}#mermaid-svg-m50KfoYxo9Y4mDvI .disabled text{fill:#efefef;}#mermaid-svg-m50KfoYxo9Y4mDvI .section-9 rect,#mermaid-svg-m50KfoYxo9Y4mDvI .section-9 path,#mermaid-svg-m50KfoYxo9Y4mDvI .section-9 circle,#mermaid-svg-m50KfoYxo9Y4mDvI .section-9 polygon,#mermaid-svg-m50KfoYxo9Y4mDvI .section-9 path{fill:hsl(180, 100%, 76.2745098039%);}#mermaid-svg-m50KfoYxo9Y4mDvI .section-9 text{fill:black;}#mermaid-svg-m50KfoYxo9Y4mDvI .node-icon-9{font-size:40px;color:black;}#mermaid-svg-m50KfoYxo9Y4mDvI .section-edge-9{stroke:hsl(180, 100%, 76.2745098039%);}#mermaid-svg-m50KfoYxo9Y4mDvI .edge-depth-9{stroke-width:-13;}#mermaid-svg-m50KfoYxo9Y4mDvI .section-9 line{stroke:hsl(0, 100%, 86.2745098039%);stroke-width:3;}#mermaid-svg-m50KfoYxo9Y4mDvI .disabled,#mermaid-svg-m50KfoYxo9Y4mDvI .disabled circle,#mermaid-svg-m50KfoYxo9Y4mDvI .disabled text{fill:lightgray;}#mermaid-svg-m50KfoYxo9Y4mDvI .disabled text{fill:#efefef;}#mermaid-svg-m50KfoYxo9Y4mDvI .section-10 rect,#mermaid-svg-m50KfoYxo9Y4mDvI .section-10 path,#mermaid-svg-m50KfoYxo9Y4mDvI .section-10 circle,#mermaid-svg-m50KfoYxo9Y4mDvI .section-10 polygon,#mermaid-svg-m50KfoYxo9Y4mDvI .section-10 path{fill:hsl(210, 100%, 76.2745098039%);}#mermaid-svg-m50KfoYxo9Y4mDvI .section-10 text{fill:black;}#mermaid-svg-m50KfoYxo9Y4mDvI .node-icon-10{font-size:40px;color:black;}#mermaid-svg-m50KfoYxo9Y4mDvI .section-edge-10{stroke:hsl(210, 100%, 76.2745098039%);}#mermaid-svg-m50KfoYxo9Y4mDvI .edge-depth-10{stroke-width:-16;}#mermaid-svg-m50KfoYxo9Y4mDvI .section-10 line{stroke:hsl(30, 100%, 86.2745098039%);stroke-width:3;}#mermaid-svg-m50KfoYxo9Y4mDvI .disabled,#mermaid-svg-m50KfoYxo9Y4mDvI .disabled circle,#mermaid-svg-m50KfoYxo9Y4mDvI .disabled text{fill:lightgray;}#mermaid-svg-m50KfoYxo9Y4mDvI .disabled text{fill:#efefef;}#mermaid-svg-m50KfoYxo9Y4mDvI .section-root rect,#mermaid-svg-m50KfoYxo9Y4mDvI .section-root path,#mermaid-svg-m50KfoYxo9Y4mDvI .section-root circle,#mermaid-svg-m50KfoYxo9Y4mDvI .section-root polygon{fill:hsl(240, 100%, 46.2745098039%);}#mermaid-svg-m50KfoYxo9Y4mDvI .section-root text{fill:#ffffff;}#mermaid-svg-m50KfoYxo9Y4mDvI .section-root span{color:#ffffff;}#mermaid-svg-m50KfoYxo9Y4mDvI .section-2 span{color:#ffffff;}#mermaid-svg-m50KfoYxo9Y4mDvI .icon-container{height:100%;display:flex;justify-content:center;align-items:center;}#mermaid-svg-m50KfoYxo9Y4mDvI .edge{fill:none;}#mermaid-svg-m50KfoYxo9Y4mDvI .mindmap-node-label{dy:1em;alignment-baseline:middle;text-anchor:middle;dominant-baseline:middle;text-align:center;}#mermaid-svg-m50KfoYxo9Y4mDvI :root{--mermaid-font-family:"trebuchet ms",verdana,arial,sans-serif;} 微生物组中介分析基准
背景
RA非AA机制尺度
组成性致假阳性
缺系统误差校准基准
方法
AA模板重采样
九法中类与类群检验
CAMRA重建AA证据
HDMT与SBMH多重检验
结果
多法FDR膨胀
不交路径全局检验失真
CAMRA校准更优
真实数据大集合存疑
意义
AA尺度中介更可解释
全局显著不等于类群中介
开源基准促标准化

背景介绍

人类肠道微生物组日益被视为联系环境暴露与宿主健康结局的关键生物中介。饮食、药物与污染物可重塑微生物群落，后者再通过代谢与免疫通路影响表型与疾病风险。现代中介分析（mediation analysis）可量化暴露经微生物组传递给结局的效应，并优先识别可能作为干预靶点的类群。

现有微生物组中介方法分为两类。其一是群落水平（community-level）全局中介检验 ，如 MedTest、PERMANOVA-med、CMM、LDM-med 全局检验、PhyloMed，输出单一综合 p 值但无法定位具体类群。其二是分类单元水平（taxon-level）中介发现，如 LDM-med、microHIMA、MarZIC、CRAmed、multimedia、SparseMCMM，旨在识别个体中介类群并给出类群水平显著性。

核心方法学难题在于：中介分析依赖两条回归------暴露→微生物组、微生物组→结局；而测序得到的是 RA，受"各类群比例之和为 1"的单元和约束，会使类群间耦合，在两条通路上都产生伪关联。若生物学相关尺度是 AA，则 RA 水平中介可能给出与机制不符的广泛"中介者名单"。作者因此首次系统基准测试群落水平与类群水平方法的错误控制，并提出面向 AA 尺度中介的 CAMRA。

重要结果

AA 数据模拟的现实性评估

已有差异丰度基准研究强调：公平的方法比较需要现实模拟。AA 靶向模拟不能只匹配 RA 边际分布，还必须捕捉 RA 与微生物载量（microbial load）的依赖。常用做法是先模拟 RA，再乘以独立模拟的载量得到"伪 AA（pseudo-AA）"。作者用 GALAXY/MicrobLiver 的实验恢复 AA 数据对比发现：真实数据中类群间对数比随载量系统变化，而 RA × 置换载量构造的伪 AA 几乎失去该依赖；全局 AA 摘要（各类群 log(AA) 均值）在真实数据中与载量强相关，在伪 AA 中近似不相关。因此，作者改以真实 AA 模板重采样生成零假设 AA，再导出 RA 供方法输入，使评价面对现实 AA 尺度真值。

图 2. 伪绝对丰度模拟未能重现真实绝对丰度数据的核心特征。a 图中，真实数据里类群间对数比（log-ratio）对微生物载量存在系统依赖；若将载量在样本间置换以打破 RA---载量耦合，该依赖大体消失。b 图中，真实 AA 轮廓里跨类群 mean log(AA) 与微生物载量强相关，而由 RA 乘以置换载量得到的伪 AA 中二者几乎不相关。结果表明，群落组成与微生物载量的依赖是真实 AA 数据的显著特征，而伪 AA 模拟策略无法捕捉该结构。

类群水平中介发现的 FDR 控制评估

在介导信号设置中，作者对五种类群水平方法基准测试，结果显示广泛 FDR 膨胀。LDM-med、MarZIC、multimedia 在几乎所有场景中均明显膨胀；CRAmed 有轻度膨胀但功效低；microHIMA 则持续保守，多数设置下经验 FDR 接近零且功效较低。Dominant+（90% 正效应 / 10% 负效应）会放大失校准，因为符号不平衡下组成扭曲更强。真实中介数越少（暴露相关与结局相关类群重叠小）时，RA 诱导伪信号越严重；增大样本量也不消除扭曲，部分方法在更大 n 下膨胀持平甚至更高，说明统计功效增强反而更容易检出 RA 水平伪关联。

图 3. 类群水平中介检验的错误发现率评估。各面板展示不同样本量、特征维度、真实中介数下的经验 FDR。a 为 Balanced+/- 设置，b 为 Dominant+ 设置。LDM-med、MarZIC、multimedia 在多数场景中 FDR 明显高于标称水平；microHIMA 偏保守；效应方向不平衡会加剧多种方法的失校准。

类群水平任意错误发现率与全局检验Ⅰ类错误

在中介零假设下（真中介集合为空），任意错误发现率（any-false-discovery rate）结果呼应信号设置：microHIMA 近零；CRAmed 在完全零与单路径零下较稳，但不交路径零设置下轻度膨胀；MarZIC 多处中度膨胀；multimedia 近全场景膨胀；LDM-med 在完全/单路径零下稳，不交路径零下严重膨胀。

全局检验的 QQ 图显示：CMM 在所有零设置下Ⅰ类错误膨胀；完全零下 LDM-med 全局、MedTest、MODIMA、PERMANOVA-med 偏保守。关键在于不交路径零（disjoint-path null）------暴露→微生物组与微生物组→结局关联都存在，但发生在不同类群------所有全局方法均大幅Ⅰ类错误膨胀，小 p 值富集、QQ 点明显偏离对角线，Dominant+ 下更甚。这说明"全局中介显著"并不等于"具体类群在 AA 尺度上中介"。

图 4. 全局中介检验在各类零设置下的 p 值分位---分位（QQ）图。完全零下多数方法偏保守；路径特指零下偏离减弱；不交路径零下所有方法均出现显著向上偏离，小 p 值过多，表明复合零结构下全局检验严重Ⅰ类错误膨胀。

补救方法：CAMRA 的构建与两种实现

CAMRA 仍以 RA 为输入，但目标为 AA 尺度中介效应。对类群 k，AA 水平暴露效应为 α_k，结局效应为 β_k，中介效应为 α_kβ_k，中介零 H0k: α_kβ_k=0 是复合零（含 (0,0)、(α≠0,β=0)、(α=0,β≠0) 三类）。CAMRA 分三步：第一步用 PALM （准泊松回归）从 RA 推 AA 水平暴露→微生物组证据，并用基于中位数的校正消去组成偏移，得 p_αk；第二步用 PALAR （变换对数比回归 + 去偏 Lasso）从 RA 推 AA 水平微生物组→结局证据，得 p_βk；第三步在复合零下做多重检验，提供两种实现：CAMRA-HDMT （估计三类子零混合比例，校准 max(p_α,p_β)）与 CAMRA-SBMH（先建路径候选集，再对双路径有据类群做子集校正中介 p 值并 FDR 调整）。

图 5. 类群水平中介发现的 FDR---功效权衡。柱高为 Power@FDR≤0.05，即在经验 FDR 不超过 0.05 的所有操作阈值中可取得的最大功效。a 为 Balanced+/-，b 为 Dominant+；横轴为真中介数 3/5/7/9，行列为样本量与特征数。CAMRA 两种变体在保持校准的同时，功效明显高于现有类群水平方法。

CAMRA 的校准、功效与计算时间

在信号设置中，CAMRA-HDMT 与 CAMRA-SBMH 均比现有类群方法显著改善校准；CAMRA-SBMH 更保守、功效略低，CAMRA-HDMT 功效更高。用 Power@FDR≤0.05 比较，CAMRA 明显占优。小 n 大 p 压力测试（p=800，n=100/200）中，CAMRA 仍保持更好校准与有竞争力功效。零设置下 CAMRA-HDMT 偶有轻微膨胀，CAMRA-SBMH 更接近目标水平。

计算时间上，全局检验里 MODIMA 最快，PERMANOVA-med 随 n 变慢，CMM 仅测 p=200；类群检验里 CAMRA 最快------CAMRA-SBMH 中位 8.11--37.93 秒，CAMRA-HDMT 10.20--40.08 秒，而 MarZIC 与 microHIMA 最大设置中位数超 2000 秒。CAMRA 在最大评估设置下仍是数十秒级。

真实数据：国家间 BMI 差异的肠道微生物组中介

作者用 curatedMetagenomicData 做"国家（中国 vs 美国）→ 肠道微生物组 → BMI"分析，经倾向得分匹配后得到 843 样本、186 物种。所有全局检验都显著（CMM p=6.6×10⁻¹⁰，LDM-med 全局 p=1.0×10⁻⁴ 等），但如模拟所示，不交路径零也会产生全局显著，因此需看类群水平结果。

在 FDR=0.05 下，模拟中膨胀的方法返回大集合：LDM-med 22 个、MarZIC 19 个、multimedia 9 个，且彼此大量唯一；校准较好的 CAMRA-HDMT 与 microHIMA 各只识别 Acidaminococcus intestini，CRAmed 无发现。路径 p 值比较显示，LDM-med 里很多物种有跨国差异丰度，但显著"物种→BMI"关联远多于 CAMRA；CAMRA 中只有 A. intestini 同时具显著暴露→类群与类群→BMI 证据。A. intestini 可产生丙酸等非丁酸短链脂肪酸，在美国组中更丰且与 BMI 正相关，符合正向中介通路；而 LDM-med 选出的一些类群两国→类群与类群→BMI 效应异号，且 RA 层面易受组成性扭曲，机制解释困难。

图 6. 国家---BMI 肠道微生物组中介的真实数据分析。a 各方法在 FDR=0.05 下检出的中介物种数及集合重叠；模拟中 FDR 膨胀的方法在真实数据中返回大得多且互异性高的中介名单。b LDM-med 与 CAMRA 的路径水平 p 值比较；许多物种在两国间差异丰度，但 BMI 关联在 LDM-med 中远更普遍。c Acidaminococcus intestini 的相对丰度按国家展示，美国组更高，且其丰度与 BMI 正相关，支持其作为正向 AA 尺度中介的生物学可解释性。

方法学参考

#mermaid-svg-oqVGk3p3FE0OhXzH{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-oqVGk3p3FE0OhXzH .edge-animation-slow{stroke-dasharray:9,5!important;stroke-dashoffset:900;animation:dash 50s linear infinite;stroke-linecap:round;}#mermaid-svg-oqVGk3p3FE0OhXzH .edge-animation-fast{stroke-dasharray:9,5!important;stroke-dashoffset:900;animation:dash 20s linear infinite;stroke-linecap:round;}#mermaid-svg-oqVGk3p3FE0OhXzH .error-icon{fill:#552222;}#mermaid-svg-oqVGk3p3FE0OhXzH .error-text{fill:#552222;stroke:#552222;}#mermaid-svg-oqVGk3p3FE0OhXzH .edge-thickness-normal{stroke-width:1px;}#mermaid-svg-oqVGk3p3FE0OhXzH .edge-thickness-thick{stroke-width:3.5px;}#mermaid-svg-oqVGk3p3FE0OhXzH .edge-pattern-solid{stroke-dasharray:0;}#mermaid-svg-oqVGk3p3FE0OhXzH .edge-thickness-invisible{stroke-width:0;fill:none;}#mermaid-svg-oqVGk3p3FE0OhXzH .edge-pattern-dashed{stroke-dasharray:3;}#mermaid-svg-oqVGk3p3FE0OhXzH .edge-pattern-dotted{stroke-dasharray:2;}#mermaid-svg-oqVGk3p3FE0OhXzH .marker{fill:#333333;stroke:#333333;}#mermaid-svg-oqVGk3p3FE0OhXzH .marker.cross{stroke:#333333;}#mermaid-svg-oqVGk3p3FE0OhXzH svg{font-family:"trebuchet ms",verdana,arial,sans-serif;font-size:16px;}#mermaid-svg-oqVGk3p3FE0OhXzH p{margin:0;}#mermaid-svg-oqVGk3p3FE0OhXzH .label{font-family:"trebuchet ms",verdana,arial,sans-serif;color:#333;}#mermaid-svg-oqVGk3p3FE0OhXzH .cluster-label text{fill:#333;}#mermaid-svg-oqVGk3p3FE0OhXzH .cluster-label span{color:#333;}#mermaid-svg-oqVGk3p3FE0OhXzH .cluster-label span p{background-color:transparent;}#mermaid-svg-oqVGk3p3FE0OhXzH .label text,#mermaid-svg-oqVGk3p3FE0OhXzH span{fill:#333;color:#333;}#mermaid-svg-oqVGk3p3FE0OhXzH .node rect,#mermaid-svg-oqVGk3p3FE0OhXzH .node circle,#mermaid-svg-oqVGk3p3FE0OhXzH .node ellipse,#mermaid-svg-oqVGk3p3FE0OhXzH .node polygon,#mermaid-svg-oqVGk3p3FE0OhXzH .node path{fill:#ECECFF;stroke:#9370DB;stroke-width:1px;}#mermaid-svg-oqVGk3p3FE0OhXzH .rough-node .label text,#mermaid-svg-oqVGk3p3FE0OhXzH .node .label text,#mermaid-svg-oqVGk3p3FE0OhXzH .image-shape .label,#mermaid-svg-oqVGk3p3FE0OhXzH .icon-shape .label{text-anchor:middle;}#mermaid-svg-oqVGk3p3FE0OhXzH .node .katex path{fill:#000;stroke:#000;stroke-width:1px;}#mermaid-svg-oqVGk3p3FE0OhXzH .rough-node .label,#mermaid-svg-oqVGk3p3FE0OhXzH .node .label,#mermaid-svg-oqVGk3p3FE0OhXzH .image-shape .label,#mermaid-svg-oqVGk3p3FE0OhXzH .icon-shape .label{text-align:center;}#mermaid-svg-oqVGk3p3FE0OhXzH .node.clickable{cursor:pointer;}#mermaid-svg-oqVGk3p3FE0OhXzH .root .anchor path{fill:#333333!important;stroke-width:0;stroke:#333333;}#mermaid-svg-oqVGk3p3FE0OhXzH .arrowheadPath{fill:#333333;}#mermaid-svg-oqVGk3p3FE0OhXzH .edgePath .path{stroke:#333333;stroke-width:2.0px;}#mermaid-svg-oqVGk3p3FE0OhXzH .flowchart-link{stroke:#333333;fill:none;}#mermaid-svg-oqVGk3p3FE0OhXzH .edgeLabel{background-color:rgba(232,232,232, 0.8);text-align:center;}#mermaid-svg-oqVGk3p3FE0OhXzH .edgeLabel p{background-color:rgba(232,232,232, 0.8);}#mermaid-svg-oqVGk3p3FE0OhXzH .edgeLabel rect{opacity:0.5;background-color:rgba(232,232,232, 0.8);fill:rgba(232,232,232, 0.8);}#mermaid-svg-oqVGk3p3FE0OhXzH .labelBkg{background-color:rgba(232, 232, 232, 0.5);}#mermaid-svg-oqVGk3p3FE0OhXzH .cluster rect{fill:#ffffde;stroke:#aaaa33;stroke-width:1px;}#mermaid-svg-oqVGk3p3FE0OhXzH .cluster text{fill:#333;}#mermaid-svg-oqVGk3p3FE0OhXzH .cluster span{color:#333;}#mermaid-svg-oqVGk3p3FE0OhXzH 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-oqVGk3p3FE0OhXzH .flowchartTitleText{text-anchor:middle;font-size:18px;fill:#333;}#mermaid-svg-oqVGk3p3FE0OhXzH rect.text{fill:none;stroke-width:0;}#mermaid-svg-oqVGk3p3FE0OhXzH .icon-shape,#mermaid-svg-oqVGk3p3FE0OhXzH .image-shape{background-color:rgba(232,232,232, 0.8);text-align:center;}#mermaid-svg-oqVGk3p3FE0OhXzH .icon-shape p,#mermaid-svg-oqVGk3p3FE0OhXzH .image-shape p{background-color:rgba(232,232,232, 0.8);padding:2px;}#mermaid-svg-oqVGk3p3FE0OhXzH .icon-shape .label rect,#mermaid-svg-oqVGk3p3FE0OhXzH .image-shape .label rect{opacity:0.5;background-color:rgba(232,232,232, 0.8);fill:rgba(232,232,232, 0.8);}#mermaid-svg-oqVGk3p3FE0OhXzH .label-icon{display:inline-block;height:1em;overflow:visible;vertical-align:-0.125em;}#mermaid-svg-oqVGk3p3FE0OhXzH .node .label-icon path{fill:currentColor;stroke:revert;stroke-width:revert;}#mermaid-svg-oqVGk3p3FE0OhXzH :root{--mermaid-font-family:"trebuchet ms",verdana,arial,sans-serif;} 真实AA模板GALAXY或MicrobLiver重采样
生成暴露相关与结局相关类群集合
AA表转RA再加多项抽样得计数矩阵
九种方法群落与类群水平检验
信号设置算经验FDR与Power
零设置算任意错误发现率与全局QQ
CAMRA三步PALM加PALAR加复合零检验
CAMRA-HDMT或CAMRA-SBMH出q值
真实数据国家BMI中介验证

可复用要点：用实验定量载量数据做 AA 模板重采样，而非 RA×独立载量；中介真值定义为 α≠0 且 β≠0 的类群；复合零不能用普通 Sobel/联合显著检验直接处理；PALM 校正组成偏移以恢复 AA 水平暴露效应；PALAR+SparCC 估 log(AA) 协方差再做去偏 Lasso；HDMT 适合作效优先，SBMH 适合校准优先；真实应用里全局显著必须配类群水平校准解释。

总结

该研究填补了微生物组中介方法系统基准测试的空白：在 RA 层面直接检验中介时，组成性会让许多方法 FDR 或Ⅰ类错误膨胀，且不交路径零会使全局检验"假阳性地显著"。作者用锚定真实 AA 的模拟证明，大中介名单未必是机制发现，可能是 RA 尺度伪关联。

CAMRA 的价值不在于"更多发现"，而在于"更可解释的发现"：它从标准 RA 输入重建 AA 意识的两路径证据，再用复合零校准做 FDR 控制，使分类单元水平中介更贴近绝对丰度机制尺度。研究也给出两条实践结论------群落水平中介显著不能当作具体类群中介的证据；类群中介方法应以校准视角评估，若有 qPCR/内标载量信息更佳，无载量时可用 CAMRA 类 AA 靶向方法。

局限在于模拟仍限于特定结局模型与加性信号，未覆盖暴露---中介交互、非线性、系统发育聚类中介等；部分方法仅用默认实现。作者开源 miMediation 与基准流程，便于社区推进微生物组中介分析的标准化。

参考文献

Wang, Q., Li, Y., Peng, Y., & Tang, Z.-Z. (2026). Error control for microbiome mediation analysis: benchmarking and remedy. Microbiome. https://doi.org/10.1186/s40168-026-02527-1
Wei, Z. et al. Fast and reliable association discovery in large-scale microbiome studies and meta-analyses using PALM. Genome Biology (2026, accepted).
Li, Y. et al. PALAR: Estimation of absolute abundance effects in regression with relative abundance predictors. Journal of the American Statistical Association (2025).
Dai, J. Y., Stanford, J. L., & LeBlanc, M. A multiple-testing procedure for high-dimensional mediation hypotheses. JASA 117(537):198--213 (2022).
Sampson, J. N. et al. FWER and FDR control when testing multiple mediators. Bioinformatics 34(14):2418--2424 (2018).
Vandeputte, D. et al. Quantitative microbiome profiling links gut community variation to microbial load. Nature 551:507--511 (2017).
CAMRA / miMediation 代码：https://github.com/tangzheng1/miMediation
基准脚本：https://github.com/tangzheng1/miMediation_benchmark
curatedMetagenomicData：https://github.com/waldronlab/curatedMetagenomicData