4 December 2015

No Succes Applying Multivariate Data Analysis?

Having difficulties using Multivariate Analysis on your data?
Many of the popular Multivariate methodsĀ  for LC/MS and GC/MS data processing, like PCA and variants of PLS find their origin in Near Infrared and Infrared Spectroscopy. Data from these techniques are highly correlated. Due to the very high correlation, PCA and regression methods like PLS often only need a few factors to solve the problem.
But this is most of the time not true for LC/MS and GC/MS data. In general peaks don’t have high correlations and if real discriminating peaks are very small, these algorithms will not be able to find them. Major peaks will always dominate the solution.

Do you recognize these problems? Try out the special algorithms from MsMetrix. We define data analysis on three levels:

  • Level 1: large peaks are responsible for separating classes/groups. No need to use multivariate tools.
  • Level 2: the problem can be solved by unique but (very) small peaks. MsCompare can find these directly at any level.
  • Level 3: the problem is really multivariate in nature. You will need to use combinations of peaks to solve the problem. Use PCA, PLS, ECVA or other multivariate methods.

On level 3, many people try to interpret multivariate solutions in a univariate manner. This can be tricky, there are limitations and restrictions to this way of interpretation.