Multivariate analysis

If we want to measure one feature, for example a concentration such as pH, we optimise our analytical method for that specific target, and get a figure corresponding to the information that we wish to obtain. However, if we want to measure more complicated features, such as how toxic a mixture of chemicals is, the progression of a disease or a medical diagnosis, it may be difficult to describe this by only one figure. This can only be described by a combination of several parameters, a multiple of variables.

As an example, the model in Figure 1 shows that only one protein concentration will not be sufficient to describe if a person is healthy or sick. However, the combined information that the concentration of protein B is low and protein A is high will indicate a sick person in this very simplified and made-up example. It is easy to imagine that a medical state in real life must be described by many more protein concentrations in combination.

Figure 1

To be able to extract the information of which variables (protein concentrations in Figure 1) that can describe our response (medical state in Figure 1), we need multivariate data analysis methods. As the name indicates, this is data analysis including more than one variable. Several multivariate statistical methods are available, some of them recently developed because they require the computational capabilities of modern computers. Generally, the multivariate methods will extract the significant variables affecting the response we are looking for. Most often this is only a limited part of all the ingoing variables.