15. Principle Components Analysis in R2

How to identify patterns or groups in your dataset using Principle Component Analysis.

15.1. Scope

  • In this tutorial expression data of a set of Medulloblastoma tumors will be investigated for the existence of subgroups.
  • Principle Component Analysis (PCA) will be used to analyze the tumor samples.

15.2. Step 1: Selecting data and modules

  1. Make sure that the Single Dataset option is selected in field 1 of the step by step guide.

  2. In field 2 locate and select the ‘Tumor Medulloblastoma PLoS One- Kool - 62 MAS5.0 -u133p2’ dataset by clicking ‘Change Dataset’

  3. In field 3 select the ‘Principle Component Option’ option.

    Figure 1: Selecting Principe Component Analysis

    Figure 1: Selecting Principe Component Analysis

  4. Click “next”

15.3. Step 2: Exploring the principle components

  1. The next window displays a set of fields where specific settings of the clustering algorithm used can be set. Leave all the settings at their default and click “next”.

  2. Click to plot the PCA result.

  3. You now see a plot of the of the first 2 principle components. In the adjustable settings box, all the combinations principle components can be selected.

  4. In the adjustable setting box select the all PCA-components option to view the several principle components combinations to investigate whether you can distinguish subgroups in your dataset.

    Figure 2: Adjusting PCA settings

    Figure 2: Adjusting PCA settings

    Figure 3: Select tracks

    Figure 3: Select Tracks

    Figure 4: Select tracks

    Figure 4: PCA components

In this example the samples are colored by known groups and fitted with the PCA result. In Figure 4 a clear subgroup, the yellow wnt subgroup is revealed. Hovering over the data points provides the principle component vector #:values which are depicted, as well as additional sample information. This example illustrated that PCA is powerful tool aiding to find possible subgroups in your dataset of interest. Also note the variance reported on the axes.

_images/R2d2_logo.pngDid you know that PCA clustering is a method that reduces data dimensionality?

Principle Component Analysis is a method that reduces data dimensionality by performing co-variance analysis between factors. PCA is especially suitable for datasets with many dimensions, such as a microarray experiment where the measurement of every single gene in a dataset can be considered a dimension. It is impossible to make a visual representation of the relation between genes and their conditions in multi-dimensional matrix. One way to make sense of data is to reduce dimensionality. Several techniques can be used for this purpose and PCA is one of them. The reduction of dimensions is archived by plotting points in a multidimensional space onto a space with fewer dimensions. The reduction is accomplished by identifying directions, so called principle components, that describe maximal variation in the data. These principle components can then be used as surrogates to represent each sample, making it possible to visually assess similarities and differences between samples and determine whether samples can be grouped. As the principle components are uncorrelated, they may represent different aspects of the samples and is therefore a powerful tool to identify subgroups in you dataset.

15.4. Step 3: Viewing clusters in 3D

A very nice feature of the R2 PCA module is the possibility to investigate your data in an interactive 3D-plotted graph. Most recent internet browsers support the 3D visualization.

_images/R2d2_logo.pngDid you know that browser settings might have to be adapted?

  1. In the adjustable settings menu select the “3d” option and click “next”.

  2. Click the cube and hold the left mouse button and rotate the picture in order to investigate whether there are any (more) subgroups that become visible.

    'Figure 6: Showing a 3D PCA graph from different angles.

    ‘Figure 6: Showing a 3D PCA graph from different angles.

  3. By rotating the graph more subgroups could be revealed as clearly shown in Figure 6.

15.5. Step 3: Using toplister for PCA

Low expression genes can significantly affect the PCA results. This has several reasons, low expressed genes give rise to stochastic noise. Many zero counts or almost zero expression values can distort the distance matrix (e.g. Euclidean and in addition there is no contribution to meaningful variability and do not reflect actual biology.

To avoid the influence for low-expressed genes or the variance between low-expressed genes values the top-lister function wih cut-offs can applied first before running the PCA algorithm as illustrated in figure 7. First generate e.g a top 1000 gene with custom cut-offs, store the generated top-1000 genes and use this list a filter.

'Figure 7: Showing a 3D PCA graph from different angles.

Figure 7: Showing a 3D PCA graph from different angles.

In the near future the cut-off function will be incorporated directly in the PCA module.

15.6. Final remarks / future directions

The identification of medulloblastoma subtypes has been published here:

Kool M, Koster J, Bunt J, Hasselt NE, Lakeman A, van Sluis P, Troost D, Meeteren NS, Caron HN, Cloos J, Mrsic A, Ylstra B, Grajkowska W, Hartmann W, Pietsch T, Ellison D, Clifford SC, Versteeg R.; Integrated genomics identifies five medulloblastoma subtypes with distinct genetic profiles, pathway signatures and clinicopathological features. PLoS One. 2008 Aug 28;3(8):e3088.

Everything described in ths chapter can be performed in the R2: genomics analysis and visualization platform (http://r2platform.com / http://r2.amc.nl)

We hope that this tutorial has been helpful, the R2 support team.