Schloss Patrick D
Department of Microbiology, University of Massachusetts--Amherst, Amherst, MA 01003, USA.
ISME J. 2008 Mar;2(3):265-75. doi: 10.1038/ismej.2008.5. Epub 2008 Jan 31.
As microbial ecology investigations have progressed from descriptive characterizations of a community to hypothesis-driven ecological research, a number of different statistical techniques have been developed to describe and compare the structure of microbial communities. Thus far, these methods have only been evaluated using 16S rRNA gene sequence data obtained from incomplete characterizations of microbial communities. In this investigation, simulations were designed to test the statistical power of different methods to differentiate between communities with known memberships and structures. These simulations revealed three important results that affect how the results of the tests are interpreted. First, integral-LIBSHUFF, TreeClimber, UniFrac, analysis of molecular variance (AMOVA) and homogeneity of molecular variance (HOMOVA) compare the structure of communities and not just their memberships. Second, integral-LIBSHUFF is unable to detect cases when one community structure is a subset of another. Third, AMOVA determines whether the genetic diversity within two or more communities is greater than their pooled genetic diversity, and HOMOVA determines whether the amount of genetic diversity in each community is significantly different. integral-LIBSHUFF, TreeClimber and UniFrac lump these and other factors together when performing their analysis making it difficult to discern the nature of the differences that are detected between communities. These findings demonstrate that when correctly employed, the current statistical toolbox has the ability to address specific ecological questions concerning the differences between microbial communities.
随着微生物生态学研究从群落的描述性表征发展到假设驱动的生态研究，已经开发了许多不同的统计技术来描述和比较微生物群落的结构。到目前为止，这些方法仅使用从微生物群落的不完整表征中获得的16S rRNA基因序列数据进行了评估。在本研究中，设计了模拟来测试不同方法区分具有已知成员和结构的群落的统计能力。这些模拟揭示了三个影响测试结果解释方式的重要结果。首先，积分-LIBSHUFF、TreeClimber、UniFrac、分子方差分析（AMOVA）和分子方差同质性（HOMOVA）比较的是群落的结构，而不仅仅是它们的成员。其次，积分-LIBSHUFF无法检测到一个群落结构是另一个群落结构的子集的情况。第三，AMOVA确定两个或更多群落内的遗传多样性是否大于它们合并后的遗传多样性，而HOMOVA确定每个群落中的遗传多样性量是否有显著差异。积分-LIBSHUFF、TreeClimber和UniFrac在进行分析时将这些因素和其他因素混在一起，使得难以辨别群落之间检测到的差异的性质。这些发现表明，当正确使用时，当前的统计工具箱有能力解决有关微生物群落差异的特定生态问题。
