Table 2. Different unsupervised feature transformation techniques and their biological findings.
This table summarizes some of the unsupervised feature transformation techniques and some relevant biological findings achieved by the methods. It also shows other methods that need some future research to assess their performance while comparing microbial communities.
| Name | Advantages | Disadvantages | Example application |
|---|---|---|---|
| Principal Component Analysis (PCA) | Allows visualization of high dimensional data using lower dimensions. | Based on Euclidean distances for dissimilarity comparisons, which can hide biologically relevant patterns. Non-linear growth in processing time. Horseshoe effect. | Showed significant correlation between relative abundance of Bacteroidetes and metagenome functions associated with obesity (4). |
| Multidimensional Scaling (MDS) / Principal Coordinate Analysis (PCoA) | Allows visualization of high dimensional data using lower dimensions allowing the use of any dissimilarity metric. | Non-linear growth in processing time. Horseshoe effect. | Showed high variability in microbial community through time while preserving differences between body sites (6). |
| Non metric Multidimensional Scaling (NMDS) | In general, preserves the high-dimensional structure with fewer axes. Based on numerical optimization, relaxing linear assumptions. | Can be more time-consuming than MDS. Arch effect. | NMDS plots showed that short-term storage conditions for soil and human related samples do not affect community composition (42). |
| FastMap | MDS approximation that relies on a mapping technique that makes linear the processing time | As this is an approximation, it might miss interesting biological patterns | Has been used in other research areas but not for microbial community comparisons |
| MetricMap | MDS approximation that expands FastMap to work on many projections at once. | As this is an approximation it might miss interesting biological patterns | Has been used in other research areas but not for microbial community comparisons |
| Landmark MDS | MDS approximation that uses a small number of landmark points to derive new coordinates. | As this is an approximation, it might miss interesting biological patterns | Has been used in other research areas but not for microbial community comparisons |