Control panel

The control panel of ClusterONE is to be found on a separate tab in the control panel of Cytoscape (see the left hand side of the Cytoscape user interface). The panel consists of two large parts: the Parameters box lets you set the parameters of the algorithm (this is subdivided into Basic parameters and Advanced parameters), while the Selection info box lets you examine some properties of the currently selected set of nodes (such as the total weight of edges within the set and at the boundary of the set, or the value of the ClusterONE quality function).

Click on the Start button at the bottom of the panel to start the clustering process. The result viewer will open automatically when the results are ready. Use the Close panel button to hide the ClusterONE control panel.

After a successful clustering process, the nodes of the network will be colored according to the number of clusters they participate in. Nodes that correspond to a single cluster only will turn red, nodes with multiple clusters will turn yellow. Gray nodes denote outliers (nodes that did not end up in any of the clusters). This coloring is implemented using a custom VizMapper style. The style is selected automatically when the clustering process finishes, but you can always return to your original style in the VizMapper panel of Cytoscape.

Parameters

The parameters are grouped into basic and advanced ones. In most of the cases, the default values of the advanced parameters should be fine, but the basic parameters may need to be adjusted to your specific needs.

Basic parameters

Minimum size
The minimum size of clusters deemed relevant by ClusterONE. This is a hard threshold: whenever ClusterONE finds a cluster smaller than the minimum size, the cluster will be discarded immediately.
Minimum density

The minimum density of clusters deemed relevant by ClusterONE. The density of a cluster is the total sum of edge weights within the cluster, divided by the number of theoretically possible edges within the cluster. In other words, this is the average edge weight within the cluster if missing edges are assumed to have a weight of zero. Whenever ClusterONE finds a cluster that has a smaller density than the value given here, the cluster will be discarded immediately. Increase the minimum density if you get too many clusters and they seem too sparse, or decrease it if you are not getting enough clusters.

Since version 0.93, ClusterONE also includes an automatic density threshold selection method, which was tuned on various protein-protein interaction datasets. When the "Auto" value is selected in the minimum density field (this can be selected by trying to choose a density threshold smaller than zero), ClusterONE will use a density threshold of 0.3 for weighted networks and 0.5 for unweighted networks. If the network is unweighted and the global clustering coefficient is smaller than 0.1, ClusterONE will assume that the network was determined from low-throughput experimental data (e.g., yeast-two-hybrid experiments) and use a higher threshold of 0.6 to compensate for the presence of many tree-like subgraphs in the input.

Edge weights
A numeric edge attribute to be used for the edge weights. [unweighted] means that each edge will have a weight equal to 1. If you don't see the name of the attribute in the list, click on the Refresh button (refresh) next to the combo box to re-scan the network for numeric edge attributes. This is necessary when you added the edge attribute you are looking for after you opened the ClusterONE control panel.

Advanced parameters

If you do not see these parameters in the control panel, click on the Advanced parameters label to expand the container holding them.

Node penalty
Penalty value corresponding to each node. When you set this option to a specific value x, ClusterONE will assume that each node has an extra boundary weight of x when it considers the addition of the node to a cluster (see [1] for more details). It can be used to model the possibility of uncharted connections for each node, so nodes with only a single weak connection to a cluster will not be added to the cluster as the penalty value will outweigh the benefits of adding the node. The default penalty value is 2.
Merging method, Overlap threshold and Similarity function

After an initial set of clusters are found, ClusterONE tries to merge highly overlapping (and thus redundant) clusters in order to clean up the result. For each pair of clusters found, ClusterONE calculates a score that quantifies the overlap between them, and two clusters are merged if this overlap is larger than a given threshold (specified by the Overlap threshold textbox). There are four different ways to calculate the overlap score, as controlled by the Similarity function combobox:

  • The match coefficient takes the size of the overlap squared, divided by the product of the sizes of the two clusters being considered, as in the paper of Bader and Hogue [2].
  • The Simpson coefficient divides the size of the overlap by the size of the smaller cluster.
  • The Jaccard similarity divides the size of the overlap by the size of the union of the two clusters.
  • The Dice similarity divides twice the size of the overlap by the sum of the sizes of the two clusters.

Merging can be done in two different ways, as controlled by the Merging method combobox:

  • The single-pass method calculates similarity scores between all pairs of complexes and creates a graph where the nodes are the complexes and two nodes are connected if the corresponding complexes have a score higher than the overlap threshold. Complexes in the same connected component of the graph will then be merged.
  • The multi-pass method calculates similarity scores between all pairs of complexes and stores those pairs that have a score larger than the overlap threshold. The highest scoring pair is then merged and the similarity of the merged complex towards its neighbors is re-calculated. This is repeated until there are no more highly overlapping complexes in the result.

The default settings (match coefficient with a threshold of 0.8 using the single-pass algorithm) seem to be satisfactory for most use-cases Decreasing the threshold will result in more clusters being merged.

Seeding method

ClusterONE works by growing clusters from initial "seeds", driven by a goal function that is maximized greedily (see the Cluster ONE paper [1] for more details). A seed can be an arbitrary subgraph, but in most cases, it is either a single node or a single edge. The seeding method prescribes how the seeds are selected during the calculation:

  • From every node means that every node will be used as a seed.
  • From unused nodes means that nodes will be tried in the descending order of their weights (where the weight of a node is the sum of the weights on its incident edges), and whenever a cluster is found, the nodes in that cluster will be excluded from the list of potential seeds. In other words, the node with the largest weight that does not participate in any of the clusters found so far will be selected as the next seed.
  • From every edge means that every edge will be considered once, each yielding a seed consisting of the two endpoints of the edge.

In practical use-cases, the From unused nodes and From every node methods are almost equivalent, but the former one yields a smaller number of redundant clusters.

Selection info

The quantities shown in this box always correspond to the currently selected set of nodes in the current Cytoscape view.

Number of nodes
The number of nodes in the cluster.
In-weight
The sum of the weights of the edges that lie completely within the cluster.
Out-weight
The sum of the weights of the edges where one endpoint lies within the cluster and the other lies outside.
Density
The density of the cluster; that is, the sum of the edge weights within the cluster divided by the number of theoretically possible edges.
Quality
The quality of the cluster, as measured by the in-weight divided by the sum of the in-weight and the out-weight. The rationale behind this measure is that a good cluster contains many heavyweight edges within the cluster itself, and it is connected to the rest of the network only by a few lightweight edges. If the penalty value set in the Advanced parameters panel is larger than zero, the out-weight of the cluster will be increased by the penalty value times the number of nodes in the cluster accordingly.
P-value
The p-value of a one-sided Mann-Whitney U test performed on the in-weights and out-weights of the vertices. A low p-value means that the in-weights are significantly larger than the out-weights, so it is more likely that the cluster is a valid finding and not the result of random fluctuations. Color codes also help distinguishing significant results from insignificant ones: p-values less than 0.05 are denoted by red colors and p-values between 0.05 and 0.1 are shown in yellow.

Note that the Mann-Whitney U test cannot be performed sensibly when all the in-weights and out-weights are equal. In this case, the p-value will show "NA".