Imagine trying to compare the microbial ecosystems of dozens of different environments. For each sample, you sequence the DNA and identify the abundances of thousands of distinct microbial species. You are left with a massive dataset where each sample is a single point existing in a space with thousands of dimensions—one dimension for every unique organism found.
Human intuition breaks down in high-dimensional spaces. We cannot visualize a thousand axes simultaneously to see if samples from similar environments cluster together. To solve this, ecologists and microbiologists rely on a powerful pipeline: converting raw composition data into pairwise distances (beta diversity), and then compressing those distances into a interpretable two- or three-dimensional map using ordination techniques like Principal Coordinate Analysis (PCoA) and Non-metric Multidimensional Scaling (NMDS).
Step 1: Compressing Composition into Distance
The first step in understanding community variation is moving away from raw species counts. Instead of asking exactly how many individuals of Species $X$ are present, we ask a simpler question: How different is Sample A from Sample B?
By calculating the dissimilarity between every possible pair of samples, we transform our complex sample-by-species matrix into a symmetric, manageable sample-by-sample distance matrix. The choice of how to calculate this distance depends heavily on the ecological question:
- Bray-Curtis Dissimilarity: This metric calculates differences based on the abundance of shared and unshared species. It is highly sensitive to dominant species. For instance, Hang Jiang et al. utilized Bray-Curtis distances to map how soil alkalinity shifts the composition of fungal communities in the wheat rhizosphere, while Heqing Huang et al. used it to contrast the bacterial profiles of river water versus river sediment.
- UniFrac Distance: Used primarily in microbiome research, UniFrac incorporates evolutionary relationships. Instead of treating all species as equally distinct, it measures the fraction of evolutionary tree branches that are unique to one sample versus another. Unweighted UniFrac focuses purely on the presence or absence of lineages, which allowed Yimeng Li et al. to identify clear structural divergences between the gut microbiotas of healthy and diarrheic musk deer.
Step 2: Projecting Distances onto a Map
Once we have a matrix of pairwise distances, we need to project these relationships onto a low-dimensional plot (typically 2D or 3D) so that spatial proximity on the screen reflects ecological similarity in real life. Two primary mathematical strategies achieve this.
Principal Coordinate Analysis (PCoA)
PCoA (also known as classical multidimensional scaling) treats the pairwise distances as literal, physical distances. Imagine a table listing the driving distances between 50 cities; PCoA uses eigenvalue decomposition to find the coordinates that recreate a geographical map of those cities on a flat piece of paper.
In a PCoA plot, the first axis (Principal Coordinate 1) represents the direction of greatest variation in the community, and the second axis represents the next orthogonal direction of variation. This linear projection is highly efficient and preserves the exact metric properties of the distance matrix. For example, Yuning Zhang et al. used PCoA to demonstrate how pelleted versus non-pelleted high-grain diets cause sheep gut microbiomes to diverge along distinct geometric trajectories.
Non-metric Multidimensional Scaling (NMDS)
Ecological data can be incredibly noisy, with massive spikes in some species and complete absences in others. In these cases, preserving literal metric distances can distort the visualization. NMDS offers a non-parametric alternative: it ignores the absolute distance values and focuses entirely on their rank order.
If Sample A is closer to Sample B than it is to Sample C in the high-dimensional space, NMDS attempts to find a low-dimensional arrangement where that rank relationship holds true. It does this iteratively, shuffling the points around on a 2D plane to minimize a metric called "stress"—a measure of how much the low-dimensional rank orders mismatch the original data.
Because it relaxes the constraint of strict linearity, NMDS is exceptionally robust for tracking complex ecological successions over time or across sharp environmental gradients. This capability is highlighted by Akira Matsui et al., who used NMDS to visualize the multi-year "ecological catch-up" of seaweed communities on artificial reefs compared to natural climax communities. Similarly, Xueli Hu et al. combined both PCoA and NMDS to map out overlapping and distinct dysbiosis patterns across multiple inflammatory diseases.
Step 3: Validating Clusters Statistically
While ordination plots provide immediate visual intuition—allowing researchers to see at a glance if samples group by geographic location, disease state, or diet—visual clustering can be deceptive. Human eyes are prone to finding patterns even in random noise. Therefore, visual ordination must be paired with multivariate statistical testing.
The standard tool for this is PERMANOVA (Permutational Multivariate Analysis of Variance). Unlike traditional ANOVA, which tests for differences between group means along a single variable, PERMANOVA analyzes the entire distance matrix. It calculates the pseudo-F statistic to compare the variance between predefined groups to the variance within those groups. To determine significance without assuming a normal distribution, it randomly shuffles (permutates) the sample labels thousands of times to see how likely the observed group separation would occur by chance.
This combination of visualization and statistical validation allows researchers to make definitive claims about ecosystem drivers. For instance, by pairing ordination with PERMANOVA, Deanna S. Beatty et al. proved that eelgrass microbiomes shift predictably in response to wasting disease across vast geographic scales. In another application, Muhammad Farooq et al. used NMDS alongside PERMANOVA to confirm that stream altitude and water chemistry are the primary forces structuring sensitive mayfly assemblages.
By converting high-dimensional biological complexity into intuitive spatial geometry, ordination techniques allow us to discover the hidden environmental and clinical forces that shape living communities.
Key Foundational Work
Tracking and modeling focus of attention in meetings (2002) — contributed 1 concept elements to this direction.
Colonic microbiome is altered in alcoholism (2012) — contributed 1 concept elements to this direction.