Skip to navigation Skip to content


Clustering of PCA results using TRENDanalysis


  • The clustering menu of TRENDanalysis is used to visualize and cluster PCA results calculated by by Trendmain or TREND NMR
  • When the readparm option is turned on (by choosing Yes), TRENDanalysis reads PC values calculated by Trendmain or TREND NMR. If readparm is turned off (by choosing No), the data will be read instead from the file specified in the pcmatrix field.
  • biplot_option indicates which PCA biplots to display. Choices are 2D3DBoth 2D and 3D, and None.
  • cluster_method provides options of clustering algorithms, The default is K-means. Parameters needed are listed in parentheses. See details under Clustering methods
  • parameter sets the parameter for clustering. See cluster_method and help_tab for details.
  • help_tab
    By default help_tab is set as Run data clustering so that TRENDanalysis will run display one or two biplots once Start is pressed. . Other options trigger listing of the parameters relevant to the cluster_method selected. Pressing start will launch the help file in html format.
  • pcn
    In this field, please list your choice of components to appear in the biplot, most often 1-3. But could be another combintaion, such as 2,3,5. At least 3 components must be selected. If 4 or more are selected, only the first 3 components will be plotted in 2D or 3D biplots. However, clustering results will be calculated using 3 or more components selected.
    The syntax is equivalent to specifying pages in a print dialog. For example, 1, 3-5, 7 means selectinig components 1, 3, 4, 5, 7.
  • scale controls scaling of output biplots. If uniform is chosen then in 2D or 3D biplots the axes will use the same scale, otherwise each axis will be scaled automatically.
  • label
    when label turns the labeling of data points in biplots on or off.
  • labelfile refers to the file containg a list of labels for the points. The format of this file is identical to file.index (See the manual of trendmain for the format of fileindex.) Note the sequence in a label file must be identical to its corresponding [fileindex] file. When no file is chosen for labelfile, TRENDanalysis will use the file names to label the data points in biplots.
  • pcmatrix When readparm is turned off, pcmatrix is used to read PCA results calculated by TRENDmain or TREND NMR. The file name ends in -pc.txt.
  • export set to on causes clustering results to be saved as prefix-cluster_2d.txt or prefix-cluster_3d.txt
  • plot
    When this option is checked, 2D and/or 3D biplots will be saved as images in the .png file format.

Choice of clustering method:

  • K-means (number of clusters)

The K-means algorithm requires the user to specify the number of clusters in parameter option. By default it is set as 3. See K-means for details.

  • Agglomerative (number of clusters)

Similar to K-means, agglomerative clustering also needs the user to specify the number of clusters. See Agglomerative Clustering for details.

  • Affinity Propagation (preference)

Affinity Propagation does not require prior-knowledge of number of clusters. Instead, preference is used to choose exemplars. Choose None in the parameter filed sets preference to the median of input similarities. See affinity_propagation for details.

  • DBSCAN (min_sample)

This option performs Density-Based Spatial Clustering of Applications (DBSCAN) clustering. The min_sample parameter defines the number of samples to be considered as a core point. See DBSCAN for details.

  • Mean Shift (bandwidth)

This option applys mean shift clustering. The parameter can be set to None. See Mean Shift for details.