More docs on the ARB website.
See also index of helppages.
Last update on 25. Nov 2018 .
Main topics:
Related topics:

Cluster detection





Cluster detection searches for subtrees of the tree selected in the ARB_DIST main window, that form homologous groups of sequences.

The main prerequisite for cluster detection to work well is a good tree, preferable a tree optimized with ARB_PARSIMONY (since cluster detection uses the same distance function as ARB_PARSIMONY does).

You may control the distance calculation by selecting filter and/or weights in the ARB_DIST main window.

The following parameters define which subtrees will be reported as clusters

  • Max. distance inside each cluster (no two sequences in a cluster have bigger distance than specified). Specify the distance as percentage of mutations, 100 means every base differs, 0 means no base differs
  • Min cluster size (clusters below that size are ignored)

Press 'Detect clusters' to start the cluster detection..

The clusters matching the given parameters will be displayed in the list below. Each line contains the following information:

  • number of species in cluster
  • mean distance [min. - max.distance]
  • minimal bases used for distance calculation (weighted)
  • a generated cluster description

Each cluster contains one so called 'representative'. The representative is the species in the cluster with the least mean distance to all other cluster members.


Working with found clusters


You can mark the members of the currently selected cluster by clicking on the 'Mark' button. Below that button you may select whether to mark
  • all species in the cluster,
  • all species in the cluster despite the representative or
  • only the representative.

The second mode is useful when you plan to remove all but the representative from the tree.
You may also mark ALL clusters by clicking on the 'Mark all' button. This is handy to expand all cluster in the tree or to load all clusters into the sequence editor.

Auto mark

If you enable the 'Auto mark' toggle, ARB will automatically mark the cluster as soon as you select it in the list.

Selecting representative

If this option is checked, the representative species of the selected cluster will be become the ´Selected species´.

Storing intermediate results

You may store the displayed clusters by either pressing
  • 'Store selected' or
  • 'Store all'

The number of currently stored clusters will be displayed on the restore button. By pressing that button, you can restore these clusters.
Press 'Swap stored' to exchange stored clusters with displayed clusters.
Storing result will be useful to compare results of two cluster detections with different parameters.

Delete results

You can delete results using 'Delete selected' or 'Clear list'.

Cluster groups

Create groups for found clusters


The performance of the cluster detection is very sensitive to the parameters:

  • Shortly said: Big cluster size + small max.distance => faster calculation
  • A cluster size of 2 forces all sequences to be loaded. This consumes time and memory and may render the calculation impossible.
  • Opposed a minimum cluster size of 10 only loads about 20% of the sequence data (in best case), a size of 20 will only load about 10% of data.
  • The bigger the maximum allowed distance is, the more clusters will be found, hence the more has to be calculated.
  • So if you got no idea about what distance to use, start with a low distance (e.g. 0.01) and if you don't find any clusters, increase the distance stepwise.



One use case is to reduce a given tree by removing clones or very nearly related species and only keeping one of them as representative of the so formed OTU.


  • search clusters
  • examine found clusters and delete those you'd like to keep
  • uncheck 'Mark representative' and click 'Mark all'
  • in ARB_NTREE call 'Tree/Remove species from tree/Remove marked'

Another use case is to create groups.

If you choose higher values for the maximum distance allowed in found clusters and for the minimum cluster size, the found clusters might be good candidates to create groups.


Be careful when the minimum distance reported for a cluster is zero. This may have 2 reasons:

  • two sequences are identical (in filtered region)
  • one sequence is empty (in filtered region)

In the second case, the results are meaningless and the empty sequence will be used as representative (which makes no sense).

As a second indicator the min. number of base positions used for distance calculation is listed for each cluster. When this gets low or zero the result get more and more random.



No bugs known