CoChat
Sign Up
Cited by 345Gut microbiota and healthMicrobial Community Ecology and PhysiologyGenomics and Phylogenetic Studies

The DOI matches its Crossref record; 68 of the 68 references it cites were found in a registry or an index.

Analysis of microbial compositions: a review of normalization and differential abundance analysis

Open access
Lin, Huang, Peddada, Shyamal Das·2020·npj Biofilms and Microbiomes
DOI 10.1038/s41522-020-00160-wPMC PMC7710733
Open this paper in CoChat — free accountOpen PDF

Abstract

Increasingly, researchers are discovering associations between microbiome and a wide range of human diseases such as obesity, inflammatory bowel diseases, HIV, and so on. The first step towards microbiome wide association studies is the characterization of the composition of human microbiome under different conditions. Determination of differentially abundant microbes between two or more environments, known as differential abundance (DA) analysis, is a challenging and an important problem that has received considerable interest during the past decade. It is well documented in the literature that the observed microbiome data (OTU/SV table) are relative abundances with an excess of zeros. Since relative abundances sum to a constant, these data are necessarily compositional. In this article we review some recent methods for DA analysis and describe their strengths and weaknesses.

Overview

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1038/s41522-020-00160-w

Cites

68 of 68 references verified0 citation errors · 0 not checkable

Statistics

Figures

Fig. 1 Microbiome data is represented by relative abundances, thus differential abundance analysis should account for the bias introduced by across-sample variations in sampling fractions. Sampling fraction is defined as the ratio of expected abundance in a sample to the corresponding absolute abundance in the ecosystem, which could be empirically estimated by the ratio of library size to the microbial load. a Differences in sampling fractions introduce false negatives. In this toy example, the microbial load for subject A in a unit volume of the ecosystem (e.g. a unit volume of gut) is 8 (4 blue + 4 red), while for subject B it is 12 (6 blue + 6 red). However, the samples taken from subject A and B have the same library size 4 (2 blue + 2 red), same observed abundance as well as the same relative abundance of blue and red taxa. Thus, one may mistakenly conclude that the blue and red taxa are not differentially abundant between two ecosystems, which is not the case in the two ecosystems. This false negative conclusion is caused by differences in the sampling fractions in the two samples. The sampling fraction in sample A is 1/2 and for B it is 1/3. b Differences in sampling fractions introduces false positives. Consider another subject C, who has the microbial load of 16 (12 blue + 4 red) in a unit volume of ecosystem. Given the same library size in sample C (3 blue + 1 red) as sample A, one may mistakenly conclude that both blue and red are differentially abundant between ecosystems A and C, while in fact, only the blue taxon is differentially abundant. Thus a normalization method must account for differences in sampling fractions to avoid such erroneous conclusions.
Fig. 1 Microbiome data is represented by relative abundances, thus differential abundance analysis should account for the bias introduced by across-sample variations in sampling fractions. Sampling fraction is defined as the ratio of expected abundance in a sample to the corresponding absolute abundance in the ecosystem, which could be empirically estimated by the ratio of library size to the microbial load. a Differences in sampling fractions introduce false negatives. In this toy example, the microbial load for subject A in a unit volume of the ecosystem (e.g. a unit volume of gut) is 8 (4 blue + 4 red), while for subject B it is 12 (6 blue + 6 red). However, the samples taken from subject A and B have the same library size 4 (2 blue + 2 red), same observed abundance as well as the same relative abundance of blue and red taxa. Thus, one may mistakenly conclude that the blue and red taxa are not differentially abundant between two ecosystems, which is not the case in the two ecosystems. This false negative conclusion is caused by differences in the sampling fractions in the two samples. The sampling fraction in sample A is 1/2 and for B it is 1/3. b Differences in sampling fractions introduces false positives. Consider another subject C, who has the microbial load of 16 (12 blue + 4 red) in a unit volume of ecosystem. Given the same library size in sample C (3 blue + 1 red) as sample A, one may mistakenly conclude that both blue and red are differentially abundant between ecosystems A and C, while in fact, only the blue taxon is differentially abundant. Thus a normalization method must account for differences in sampling fractions to avoid such erroneous conclusions.CC BY · PubMed Central
Fig. 2 Rarefaction curves using the diet swap data68 at the genus level. The number of genera is 130, and the sample size is 222 (African American = 123, Native African = 99). x axis denotes the library size, and y axis represents the corresponding alpha diversity. Data are presented as mean values ± standard error (SE). It shows that regardless of the choice of diversity measures, as the increase of library size, the rarefaction curve starts to “level out” suggesting that the diversity of the samples has been fully observed.
Fig. 2 Rarefaction curves using the diet swap data68 at the genus level. The number of genera is 130, and the sample size is 222 (African American = 123, Native African = 99). x axis denotes the library size, and y axis represents the corresponding alpha diversity. Data are presented as mean values ± standard error (SE). It shows that regardless of the choice of diversity measures, as the increase of library size, the rarefaction curve starts to “level out” suggesting that the diversity of the samples has been fully observed.CC BY · PubMed Central
Fig. 3 Box plot of residuals between true sampling fraction and its estimate for each sample. In the box plot, the lower and upper hinges correspond to the first and third quartiles (the 25th and 75th percentiles). The median is represented by a solid line within the box. The upper whisker extends from the hinge to the largest value (maxima) no further than 1.5 times Interquartile Range (IQR, distance between the first and third quartiles) from the hinge, the lower whisker extends from the hinge to the smallest value (minima) at most 1.5 times IQR of the hinge. Data beyond the end of the whiskers are called “outlying” points. N = 90 samples examined over three study groups (denoted by circle, cross, and triangle, with 30 samples per group) and the data points are overlaid in each box. Each facet title indicates the normalization method and its variance is provided within parenthesis. The microbial absolute abundances in the ecosystem are generated from the log-normal distribution. By comparing residuals across different groups, an ideal box-plot should display a narrow height (i.e. smaller variability) and samples from different groups should be inter-mixed and not display any systematic separation. We note that all existing methods have larger variances compared to ANCOM-BC, and TSS has the largest variance. Except ANCOM-BC, UQ, and TMM, we see from the plot that circles, cross, and triangles are systematically separated, which indicates that ELib-UQ, ELib-TMM, CSS, MED, and TSS do not account for systematic bias due to differences in sampling fractions across groups.
Fig. 3 Box plot of residuals between true sampling fraction and its estimate for each sample. In the box plot, the lower and upper hinges correspond to the first and third quartiles (the 25th and 75th percentiles). The median is represented by a solid line within the box. The upper whisker extends from the hinge to the largest value (maxima) no further than 1.5 times Interquartile Range (IQR, distance between the first and third quartiles) from the hinge, the lower whisker extends from the hinge to the smallest value (minima) at most 1.5 times IQR of the hinge. Data beyond the end of the whiskers are called “outlying” points. N = 90 samples examined over three study groups (denoted by circle, cross, and triangle, with 30 samples per group) and the data points are overlaid in each box. Each facet title indicates the normalization method and its variance is provided within parenthesis. The microbial absolute abundances in the ecosystem are generated from the log-normal distribution. By comparing residuals across different groups, an ideal box-plot should display a narrow height (i.e. smaller variability) and samples from different groups should be inter-mixed and not display any systematic separation. We note that all existing methods have larger variances compared to ANCOM-BC, and TSS has the largest variance. Except ANCOM-BC, UQ, and TMM, we see from the plot that circles, cross, and triangles are systematically separated, which indicates that ELib-UQ, ELib-TMM, CSS, MED, and TSS do not account for systematic bias due to differences in sampling fractions across groups.CC BY · PubMed Central
Fig. 4 False Discovery Rate (FDR) and power comparisons using synthetic data. The FDR and power of various differential abundance (DA) analyses (two-sided) are shown in a, b, respectively. The number of taxa is set to be 200, and the sample size equals to 60, with 30 samples per group. The microbial absolute abundances in the ecosystem are generated from the log-normal distribution. The y-axis denotes patterns of proportion of differential abundant taxa, ranging from 0.05 to 0.25. The solid vertical line is the 5% nominal level of FDR, and the dashed vertical line denotes 5% nominal level plus one standard error (SE). Legend on the bottom indicates the color for each method and the normalization method is provided within parenthesis. By default, ANCOM-BC implements Holm-Bonferroni70 method and other DA methods implement BH procedure57 to adjust for multiple comparisons. By detecting differentially abundant taxa between two groups, results show that only ANCOM and ANCOM-BC control the FDR under the nominal level (5%) while maintaining power comparable to other methods. Gaussian model version of metagenomeSeq has highly inflated FDR, while the log Gaussian version has substantial loss of power, sometimes well below 5%.
Fig. 4 False Discovery Rate (FDR) and power comparisons using synthetic data. The FDR and power of various differential abundance (DA) analyses (two-sided) are shown in a, b, respectively. The number of taxa is set to be 200, and the sample size equals to 60, with 30 samples per group. The microbial absolute abundances in the ecosystem are generated from the log-normal distribution. The y-axis denotes patterns of proportion of differential abundant taxa, ranging from 0.05 to 0.25. The solid vertical line is the 5% nominal level of FDR, and the dashed vertical line denotes 5% nominal level plus one standard error (SE). Legend on the bottom indicates the color for each method and the normalization method is provided within parenthesis. By default, ANCOM-BC implements Holm-Bonferroni70 method and other DA methods implement BH procedure57 to adjust for multiple comparisons. By detecting differentially abundant taxa between two groups, results show that only ANCOM and ANCOM-BC control the FDR under the nominal level (5%) while maintaining power comparable to other methods. Gaussian model version of metagenomeSeq has highly inflated FDR, while the log Gaussian version has substantial loss of power, sometimes well below 5%.CC BY · PubMed Central
Fig. 5 Venn diagrams representing consistency of differentially abundant genera identified by ANCOM-BC, ANCOM, and DR. The global gut microbiota data60 is used to make the Venn diagram. The dataset contains 673 genera, with subjects from Malawi (MA, n1 = 114), USA (US, n2 = 317), and Venezuela (VEN, n3 = 99). We compare the absolute abundance of different genera for (1) Subjects who are less than or equal to 2 years old with sample size (MA:US:VEN) = (47:50:27) and (2) Subjects who are greater than 2 years old with sample size (MA:US:VEN) = (36:260:70). ANCOM-BC and ANCOM generally have large overlap of significant genera.
Fig. 5 Venn diagrams representing consistency of differentially abundant genera identified by ANCOM-BC, ANCOM, and DR. The global gut microbiota data60 is used to make the Venn diagram. The dataset contains 673 genera, with subjects from Malawi (MA, n1 = 114), USA (US, n2 = 317), and Venezuela (VEN, n3 = 99). We compare the absolute abundance of different genera for (1) Subjects who are less than or equal to 2 years old with sample size (MA:US:VEN) = (47:50:27) and (2) Subjects who are greater than 2 years old with sample size (MA:US:VEN) = (36:260:70). ANCOM-BC and ANCOM generally have large overlap of significant genera.CC BY · PubMed Central
Fig. 6 Waterfall plot visualizing coefficient (US: MA) for top 20 balances identified by gneiss using the global gut microbiota data60. We subset the dataset to subjects whose nationality is Malawi (MA) and USA (US), with sample size (MA:US) = (47:50) for the group of age ≤2 years old, and (MA:US) = (36:260) for the group of age >2 years old. The columns of the plot represent coefficients, and the rows of the plot represent balances. BH procedure57 was applied to correct for multiple comparisons, and coefficients with FDR corrected p-values < 0.05 are discarded.
Fig. 6 Waterfall plot visualizing coefficient (US: MA) for top 20 balances identified by gneiss using the global gut microbiota data60. We subset the dataset to subjects whose nationality is Malawi (MA) and USA (US), with sample size (MA:US) = (47:50) for the group of age ≤2 years old, and (MA:US) = (36:260) for the group of age >2 years old. The columns of the plot represent coefficients, and the rows of the plot represent balances. BH procedure57 was applied to correct for multiple comparisons, and coefficients with FDR corrected p-values < 0.05 are discarded.CC BY · PubMed Central

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

Show all 68 references

Research works printed but not linked: 1

  1. Gotelli, N. J. & Colwell, R. K. Estimating species richness. Biol. Divers. Front. Meas. Assess. 12, 39–54 (2011).

Cited by

Sorted by citations

See all 345

Related

Metadata

Authors
Lin, H., & Peddada, S. D.
Affiliations
National Institutes of Health (US)
Corresponding author
Peddada, Shyamal Das
ORCID
Venue
npj Biofilms and Microbiomes
Publisher
Springer Science and Business Media LLC
Published
2020-12-02
Volume
6
Issue
1
Pages
60-60
ISSN
2055-5008 (ISSN-L)
DOI
10.1038/s41522-020-00160-w
PMC
PMC7710733
OpenAlex
W3109541743
Cited by
341
References
68
Citations by year
2026: 57 · 2025: 73 · 2024: 67 · 2023: 77 · 2022: 55 · 2021: 11
FWCI
3.45
Type
journal article
Language
en
Open access
Licence
Locations
  • npj Biofilms and Microbiomes (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
Funders
University of Pittsburgh
Topics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Gut microbiota and health
  • Physical Sciences › Environmental Science › Ecology › Microbial Community Ecology and Physiology
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Genomics and Phylogenetic Studies
Keywords
microbiome, differential abundance, compositional data, normalization, zero-inflated data, OTU table, sequencing read counts, human microbiome
MeSH terms
Gastrointestinal Microbiome, Bacteria, Humans, Models, Theoretical, Mouth, Phylogeny

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review