Read Taxonomy Assignment - Methods

Read Taxonomy Assignment - Result Summary *

Read Taxonomy Assignment - ASV Species-Level Read Counts Table

Alpha Diversity Analysis by Rarefaction

Diversity measures are affected by the sampling depth. Rarefaction is a technique to assess species richness from the results of sampling. Rarefaction allows the calculation of species richness for a given number of individual samples, based on the construction of so-called rarefaction curves. This curve is a plot of the number of species as a function of the number of samples. Rarefaction curves generally grow rapidly at first, as the most common species are found, but the curves plateau as only the rarest species remain to be sampled.

References:
Willis AD. Rarefaction, Alpha Diversity, and Statistics. Front Microbiol. 2019 Oct 23;10:2407. doi: 10.3389/fmicb.2019.02407. PMID: 31708888; PMCID: PMC6819366.

Boxplot of Alpha-diversity Indices

The two main factors taken into account when measuring diversity are richness and evenness. Richness is a measure of the number of different kinds of organisms present in a particular area. Evenness compares the similarity of the population size of each of the species present. There are many different ways to measure the richness and evenness. These measurements are called "estimators" or "indices". Below is a diversity of 3 commonly used indices showing the values for all the samples (dots) and in groups (boxes).

Group Significance of Alpha-diversity Indices

To test whether the alpha diversity among different comparison groups are different statisticall, we use the Kruskal Wallis H test provided the "alpha-group-significance" fucntion in the QIIME 2 diversity package. Kruskal Wallis H test is the non parametric alternative to the One Way ANOVA. Non parametric means that the test doesn’t assume your data comes from a particular distribution. The H test is used when the assumptions for ANOVA aren’t met (like the assumption of normality). It is sometimes called the one-way ANOVA on ranks, as the ranks of the data values are used in the test rather than the actual data points. The test determines whether the medians of two or more groups are different.

Below are the Kruskal Wallis H test results for each comparison based on three different alpha diversity measures: 1) Observed species (features), 2) Shannon index, and 3) Simpson index.

NMDS and PCoA Plots

Beta diversity compares the similarity (or dissimilarity) of microbial profiles between different groups of samples. There are many different similarity/dissimilarity metrics. In general, they can be quantitative (using sequence abundance, e.g., Bray-Curtis or weighted UniFrac) or binary (considering only presence-absence of sequences, e.g., binary Jaccard or unweighted UniFrac). They can be even based on phylogeny (e.g., UniFrac metrics) or not (non-UniFrac metrics, such as Bray-Curtis, etc.).

For microbiome studies, species profiles of samples can be compared with the Bray-Curtis dissimilarity, which is based on the count data type. The pair-wise Bray-Curtis dissimilarity matrix of all samples can then be subject to either multi-dimensional scaling (MDS, also known as PCoA) or non-metric MDS (NMDS).

MDS/PCoA is a scaling or ordination method that starts with a matrix of similarities or dissimilarities between a set of samples and aims to produce a low-dimensional graphical plot of the data in such a way that distances between points in the plot are close to original dissimilarities.

NMDS is similar to MDS, however it does not use the dissimilarities data, instead it converts them into the ranks and use these ranks in the calculation.

In our beta diversity analysis, Bray-Curtis dissimilarity matrix was first calculated and then plotted by the PCoA and NMDS separately. Below are beta diveristy results for all groups together:

The above PCoA and NMDS plots are based on count data. The count data can also be transformed into centered log ratio (CLR) for each species. The CLR data is no longer count data and cannot be used in Bray-Curtis dissimilarity calculation. Instead CLR can be compared with Euclidean distances. When CLR data are compared by Euclidean distance, the distance is also called Aitchison distance.

Below are the NMDS and PCoA plots of the Aitchison distances of the samples:

Interactive 3D PCoA Plots - Bray-Curtis Dissimilarity

Interactive 3D PCoA Plots - Euclidean Distance

Interactive 3D PCoA Plots - Correlation Coefficients

Group Significance of Beta-diversity Indices

To test whether the between-group dissimilarities are significantly greater than the within-group dissimilarities, the "beta-group-significance" function provided in the QIIME 2 "diversity" package was used with PERMANOVA (permutational multivariate analysis of variance) chosen s the group significan testing method.

Three beta diversity matrics were used: 1) Bray–Curtis dissimilarity 2) Correlation coefficient matrix , and 3) Aitchison distance (Euclidean distance between clr-transformed compositions).

16S rRNA next generation sequencing (NGS) generates a fixed number of reads that reflect the proportion of different species in a sample, i.e., the relative abundance of species, instead of the absolute abundance. In Mathematics, measurements involving probabilities, proportions, percentages, and ppm can all be thought of as compositional data. This makes the microbiome read count data “compositional” (Gloor et al, 2017). In general, compositional data represent parts of a whole which only carry relative information (http://www.compositionaldata.com/).

The problem of microbiome data being compositional arises when comparing two groups of samples for identifying “differentially abundant” species. A species with the same absolute abundance between two conditions, its relative abundances in the two conditions (e.g., percent abundance) can become different if the relative abundance of other species change greatly. This problem can lead to incorrect conclusion in terms of differential abundance for microbial species in the samples.

When studying differential abundance (DA), the current better approach is to transform the read count data into log ratio data. The ratios are calculated between read counts of all species in a sample to a “reference” count (e.g., mean read count of the sample). The log ratio data allow the detection of DA species without being affected by percentage bias mentioned above

In this report, a compositional DA analysis tool “ANCOM” (analysis of composition of microbiomes) was used. ANCOM transforms the count data into log-ratios and thus is more suitable for comparing the composition of microbiomes in two or more populations. "ANCOM" generates a table of features with W-statistics and whether the null hypothesis is rejected. The “W” is the W-statistic, or number of features that a single feature is tested to be significantly different against. Hence the higher the "W" the more statistical sifgnificane that a feature/species is differentially abundant.

References:

Gloor GB, Macklaim JM, Pawlowsky-Glahn V, Egozcue JJ. Microbiome Datasets Are Compositional: And This Is Not Optional. Front Microbiol. 2017 Nov 15;8:2224. doi: 10.3389/fmicb.2017.02224. PMID: 29187837; PMCID: PMC5695134.

Mandal S, Van Treuren W, White RA, Eggesbø M, Knight R, Peddada SD. Analysis of composition of microbiomes: a novel method for studying microbial composition. Microb Ecol Health Dis. 2015 May 29;26:27663. doi: 10.3402/mehd.v26.27663. PMID: 26028277; PMCID: PMC4450248.

Lin H, Peddada SD. Analysis of compositions of microbiomes with bias correction. Nat Commun. 2020 Jul 14;11(1):3514. doi: 10.1038/s41467-020-17041-7. PMID: 32665548; PMCID: PMC7360769.

Starting with version V1.2, we also include the results of ANCOM-BC (Analysis of Compositions of Microbiomes with Bias Correction) (Lin and Peddada 2020). ANCOM-BC is an updated version of "ANCOM" that: (a) provides statistically valid test with appropriate p-values, (b) provides confidence intervals for differential abundance of each taxon, (c) controls the False Discovery Rate (FDR), (d) maintains adequate power, and (e) is computationally simple to implement. The bias correction (BC) addresses a challenging problem of the bias introduced by differences in the sampling fractions across samples. This bias has been a major hurdle in performing DA analysis of microbiome data. ANCOM-BC estimates the unknown sampling fractions and corrects the bias induced by their differences among samples. The absolute abundance data are modeled using a linear regression framework.

References:

Lin H, Peddada SD. Analysis of compositions of microbiomes with bias correction. Nat Commun. 2020 Jul 14;11(1):3514. doi: 10.1038/s41467-020-17041-7. PMID: 32665548; PMCID: PMC7360769.

LEfSe (Linear Discriminant Analysis Effect Size) is an alternative method to find "organisms, genes, or pathways that consistently explain the differences between two or more microbial communities" (Segata et al., 2011). Specifically, LEfSe uses rank-based Kruskal-Wallis (KW) sum-rank test to detect features with significant differential (relative) abundance with respect to the class of interest. Since it is rank-based, instead of proportional based, the differential species identified among the comparison groups is less biased (than percent abundance based).

Reference:

Segata N, Izard J, Waldron L, Gevers D, Miropolsky L, Garrett WS, Huttenhower C. Metagenomic biomarker discovery and explanation. Genome Biol. 2011 Jun 24;12(6):R60. doi: 10.1186/gb-2011-12-6-r60. PMID: 21702898; PMCID: PMC3218848.

Species vs Sample Abundance Heatmap for All Samples

Heatmaps for Individual Comparisons

To analyze the co-occurrence or co-exclusion between microbial species among different samples, network correlation analysis tools are usually used for this purpose. However, microbiome count data are compositional. If count data are normalized to the total number of counts in the sample, the data become not independent and traditional statistical metrics (e.g., correlation) for the detection of specie-species relationships can lead to spurious results. In addition, sequencing-based studies typically measure hundreds of OTUs (species) on few samples; thus, inference of OTU-OTU association networks is severely under-powered. Here we use SPIEC-EASI (SParse InversE Covariance Estimation for Ecological Association Inference), a statistical method for the inference of microbial ecological networks from amplicon sequencing datasets that addresses both of these issues (Kurtz et al., 2015). SPIEC-EASI combines data transformations developed for compositional data analysis with a graphical model inference framework that assumes the underlying ecological association network is sparse. SPIEC-EASI provides two algorithms for network inferencing – 1) Meinshausen-Bühlmann's neighborhood selection (MB method) and inverse covariance selection (GLASSO method, i.e., graphical least absolute shrinkage and selection operator). This is fundamentally distinct from SparCC, which essentially estimate pairwise correlations. In addition to these two methods, we provide the results of a third method - SparCC (Sparse Correlations for Compositional Data)(Friedman & Alm 2012), which is also a method for inferring correlations from compositional data. SparCC estimates the linear Pearson correlations between the log-transformed components.

References:

Kurtz ZD, Müller CL, Miraldi ER, Littman DR, Blaser MJ, Bonneau RA. Sparse and compositionally robust inference of microbial ecological networks. PLoS Comput Biol. 2015 May 7;11(5):e1004226. doi: 10.1371/journal.pcbi.1004226. PMID: 25950956; PMCID: PMC4423992.

Friedman J, Alm EJ. Inferring correlation networks from genomic survey data. PLoS Comput Biol. 2012;8(9):e1002687. doi: 10.1371/journal.pcbi.1002687. Epub 2012 Sep 20. PMID: 23028285; PMCID: PMC3447976.

SPIEC-EASI Network Inference by Neighborhood Selection (MB Method)

Association Network Inference by SparCC