MDS，PCoA和PCA

1，MDS

Principal Component Analysis (PCA) and classical Multidimensional Scaling (MDS) are closely related. The main difference is the type of input they take:

PCA: A dataset where each row is an observation, and each column is a feature.
MDS: A distance matrix.（这里的距离包含各种各样的距离，或者距离的rank）

MDS分为两种类型：1，metric MDS，它使用距离矩阵进行计算；2，Non-metric MDS，用距离的排序构造矩阵。

What is metric MDS?
Metric MDS is a type of MDS that assumes that the distances between the data points in the original space can be preserved in the lower-dimensional space. It uses a numerical optimization algorithm to find the best configuration of points that minimizes the discrepancy between the original and the reduced distances. Metric MDS can handle both interval and ratio data, and it can produce a stress value that measures the quality of the fit.

What is nonmetric MDS?
Nonmetric MDS is a type of MDS that does not assume that the distances between the data points in the original space can be preserved in the lower-dimensional space. Instead, it only preserves the rank order of the distances, that is, which pairs of points are closer or farther apart than others. It uses an iterative procedure to find the best configuration of points that maximizes the correlation between the original and the reduced distances. Nonmetric MDS can handle ordinal data, and it can produce a stress value and a coefficient of determination that measure the quality of the fit.

Advantages of metric MDS
One of the main advantages of metric MDS is that it can capture the linear relationships between the data points more accurately than nonmetric MDS. It can also handle negative distances, which may occur when the data points are centered or standardized. Moreover, metric MDS can be easier to interpret, since it preserves the absolute distances and scales of the original data.

Disadvantages of metric MDS
One of the main disadvantages of metric MDS is that it can be sensitive to outliers and noise in the data, which may distort the configuration of points and the dimensions. It can also be affected by the choice of distance measure, which may not reflect the true similarity or dissimilarity among the data points. Furthermore, metric MDS can be computationally intensive, especially for large data sets.

Advantages of nonmetric MDS
One of the main advantages of nonmetric MDS is that it can capture the nonlinear relationships between the data points better than metric MDS. It can also handle ordinal data, which may not have a meaningful distance measure. Moreover, nonmetric MDS can be more robust to outliers and noise in the data, since it only relies on the rank order of the distances.

Disadvantages of nonmetric MDS
One of the main disadvantages of nonmetric MDS is that it can lose some information about the magnitude and direction of the distances between the data points, which may affect the interpretation of the dimensions. It can also be affected by the choice of monotonic transformation, which may not reflect the true similarity or dissimilarity among the data points. Furthermore, nonmetric MDS can be computationally intensive, especially for large data sets and high dimensions.

2，PCoA

Non-metric的MDS就叫做Principal Coordinates Analysis（PCoA），它跟PCA的区别在于用距离矩阵代替了PCA中的观察值。