Multivariate Analyses of Microbial Communities

​

​

​

Goals of Next Two Days

​

• Explore the huge variety of statistical tests available for the interrogation of microbial community data

​

• Gain experience of choosing and using these tests within the R computing language

​

• Be aware of the limits and pitfalls of using these tests on particular microbial datasets

Microbial Communities Analysis Workflow

Raw

Sequences

Quality

Trimming

Dereplicate &

Denoise

Chimera

Detection

Taxonomy

Assignment

1

4

3

2

5

BIOINFORMATICS

DOWNSTREAM ANALYSIS

Quality

Filtering

6

Rarefied

Libraries

7

Alpha

Diversity

8

Unrarefied

Libraries

10

Microbial Communities Analysis Workflow

BIOINFORMATICS

1-5

Beta

Diversity

9

Differential

Abundance

Analysis

11

T tests, GLMs,

GLMMS

Heatmaps, PERMANOVA, Ordinations,

Divergence, Indicator Analysis

DESeq2

ALDEX2

CLR-

Transformations

Microbial Communities Are Complex

Essential Reading

https://onlinelibrary.wiley.com/doi/abs/10.1111/mec.13536

​

https://academic.oup.com/femsec/article/62/2/142/434668

​

Part 1: Alpha Diversity Analysis

What Are We Measuring?

What Are We Measuring?

Alpha diversity metrics are CORRELATED

YUNGAY

BAQUEDANO

2 TRANSECTS

MULTIPLE SITES

PER TRANSECT

YU1

YU2

YUx

BA1

BA2

BAx

MULTIPLE SAMPLES

PER SITE

Sample 1

Sample 2

Sample x

Sample 1

Sample 2

Sample x

Our data are

STRUCTURED

Samples are

not independent

data points

Part 2: Beta Diversity Analysis

Beta Diversity

We want to understand the differences in microbial community composition among samples and among groups of samples

Beta Diversity

We want to understand the differences in microbial community composition among samples and among groups of samples

But how do we summarise microbial communities with potentially hundreds of traits (species) ?

Beta Diversity

To do this, we need

​

• a DISTANCE metric (how dissimilar are samples)

​

• an ORDINATION technique (how do we display those dissimilarities in 2 (ish) dimensions

​

Ordination

Ordination methods like PCA:

​

1. Calculate new synthetic variables (principal components), which are linear combinations of the original variables that…
2. account for as much of the variance of the original data as possible

​

Ordination Techniques

EXPLORATORY ORDINATIONS:

• Principal Coordinates Analysis (PCoA): Like a Principal Component Analysis (PCA) but can use any dissimilarity matrix (Bray-Curtis, UNIFRAC) etc. Can’t handle predictors.

​

• Non-Metric Multidimensional Scaling (NMDS): Unique ordination technique in which you choose number of axes to ordinate to. Rank based approach, not absolute distances. Can’t handle predictors.

​

INTERPRETIVE ORDINATIONS

• Constrained Correspondence Analysis (CCA): Calculates axes that ‘constrain’ variation in distance matrix to that explained by candidate variables, and axes that ordinate variance not explained by those factors.

​

Ordination Metrics (Can Be) Correlated

TESTS of BETA DIVERSITY

Paliy & Shankar

TESTS OF BETA DIVERSITY

PERMANOVA

The null hypothesis tested by PERMANOVA is that, under the assumption of exchangeability of the sample units among the groups, H₀: “the centroids of the groups, as defined in the space of the chosen resemblance measure, are equivalent for all groups.” Thus, if H₀ were true, any observed differences among the centroids in a given set of data will be similar in size to what would be obtained under random allocation of individual sample units to the groups (i.e., under permutation). 

​

ANOSIM

The null hypothesis for the ANOSIM test is closely related to this, namely H₀: “the average of the ranks of within‐group distances is greater than or equal to the average of the ranks of between‐group distances,” 

PERMUTATIONAL TESTING

AXIS 1

AXIS 2

Vegetation

No Vegetation

REAL DATA

Between

Group

Within

Group

PERMUTATIONAL TESTING

AXIS 1

AXIS 2

Vegetation

No Vegetation

REAL DATA

Between

Group

Within

Group

AXIS 1

AXIS 2

PERMUTED DATA

Between

Group

Within

Group

TESTS OF BETA DIVERSITY

• Overall, ANOSIM and the Mantel test were very sensitive to heterogeneity in dispersions, with ANOSIM generally being more sensitive than the Mantel test.

​

• In contrast, PERMANOVA ... [was] largely unaffected by heterogeneity for balanced designs.

​

• For unbalanced designs, however, all of the tests were (1) too liberal when the smaller group had greater dispersion and (2) overly conservative when the larger group had greater dispersion, especially ANOSIM and the Mantel test.

​

• For simulations based on real ecological data sets, PERMANOVA was generally, but not always, more powerful than the others to detect changes in community structure

https://esajournals.onlinelibrary.wiley.com/doi/10.1890/12-2010.1

The Importance of Metadata

https://royalsocietypublishing.org/doi/abs/10.1098/rspb.2021.0552

The Importance of Metadata

The Importance of Metadata

The Importance of Metadata��Statistical Power and Data Completeness�

EXPERIMENTAL DESIGN

The Importance of Metadata��Statistical Power and Data Completeness�

DATA COLLECTION

The Importance of Metadata��Statistical Power and Data Completeness�

LAB ISSUES