A brief overview of the package

This package was created for an introductory PhD course on Chemometrics given at Section of Chemical Engineering, Aalborg University. Quickly I found out that using R for this course (with all advantages it gives) needs a lot of routine work from students, since most of them were also beginners in R. Of course it is nice when students get to know how to calculate explained variance and residuals manually, make corresponding plots and so on, but for the introductory course these things (as well as numerous typos and small mistakes in a code) take too much time, which can be spent for explaining methods and proper interpretation of results.

This is actually true for everyday using of these methods, most of the routines can be written ones and simply reused with various options. So it was decided to write a package where most widely used chemometric methods for multivariate data analysis are implemented and which gives also quick and easy-to-use access to results, produced by these methods. First of all numerous plots.

Here how it works. Say, we need to make a PCA model for data matrix x with autoscaling. Then make an overview of most important plots and investigate scores and loadings for first three components. The mdatools solution will be:

# make a model 
m = pca(x, scale = T)

#show explained variance plot
plotVariance(m)

# select optimal number of components (say, 4) for calculation of residuals
m = selectCompNum(m, 4)

# show plots for model overview
plot(m)

# show scores plot for PC1 and PC3
plotScores(m, c(1, 3))

# show loadings plot for the same components
plotLoadings(m, c(1, 3))

# show the loadings as a set of bar plots
plotLoadings(m, c(1, 3), type = 'h')

Fairly simple, is not it? The other "routine", which have been taken into account is validation — any model can be cross-validated or validated with test set and the model object will contain the validation results, which will also appear on all model plots, etc. See the next chapters for details.

It is recommended to start with the chapter about Principal Component Analysis, as I use it to explain main concepts and ideas of the package. Most of the things, explained in this chapter (such as models and results, validation, different kinds of plots, etc), work similar in other methods.

What mdatools can do?

The package includes classes and functions for analysis, preprocessing and plotting data and results. So far the following methods for analysis are implemented:

  • Principal Component Analysis (PCA)
  • Soft Independent Modelling of Class Analogy (SIMCA)
  • Partial Least Squares regression (PLS)
  • Partial Least Squares Discriminant Analysis (PLS-DA)
  • Randomisation test for PLS regression models
  • Interval PLS for variable selection

Preprocessing methods include:

  • Mean centering, standardization and autoscaling
  • Savitzky-Golay filter for smoothing and derivatives
  • Standard Normal Variate for removing scatter effect from spectral data
  • Mutliplicative Scatter Correction for the same issues
  • Normalization of spectra to unit area or unit length

More methods both for analysis and preprocessing are coming. Besides that, some extensions for the basic R plotting functionality have been also implemented and allow to do the following:

  • Color grouping of objects with automatic color legend bar.
  • Plot for several groups of objects with automatically calculated axes limits and plot legend.
  • Two built-in color schemes — one is based on a diverging scheme from Colorbrewer and the other one is a grayscale scheme.
  • Very easy-to-use possibility to apply any user defined color scheme.
  • Possibility to show horizontal and vertical lines on the plot with automatically adjusted axes limits.

See ?mdatools for more details.

How to install

The package now is available from CRAN by usual installing procedure. However due to restrictions in CRAN politics regarding number of submissions (once in 3-4 month) only major releases will be published there. To get the latest release plase use GitHub sources. You can either download a zip-file with the source package and install it using the install.packages command, e.g. if the downloaded file is mdatools_0.7.0.tar.gz and it is located in a current working directory, just run the following:

install.packages('mdatools_0.7.0.tar.gz')

If you have devtools package installed, the following command will install the latest release from the GitHub (do not forget to load the devtools package first):

install_github('svkucheryavski/mdatools')

results matching ""

    No results matching ""