See this blog for information on the most recent version of CASS.
CASS constructs a phylogenetic network, which is a phylogenetic tree with reticulations. A reticulation consists of two or more combining branches. For example, the network above contains four reticulations, which are displayed in blue. If a reticulation consists of two combining branches, then this basically indicates that one part of the genome evolved using one of these branches, while the other part of the genome evolved using the other branch.
The precise interpretation of reticulations depends on the type of data used. Possible interpretations are hybridisation, recombination or lateral gene transfer (horizontal gene transfer). However, conflicts between trees can have other causes, such as lineage sorting, gene duplication and loss and sequencing errors. These conflicts will also be resolved using reticulations. Thus, one has to interpret phylogenetic networks carefully, since not each reticulation necessarily needs to correspond to a hybridisation, recombination or horizontal gene transfer. However, the phylogenetic network can always be seen as a representation of the available data, in the sense described below.
Clusters, which are sometimes called clades or monophyletic groups, are groups of taxa that contain all descendents of their most recent common ancestor. CASS (as well as the galled and cluster network methods) constructs a phylogenetic network that represents all clusters from all input trees. If a threshold percentage t is used, then the program makes sure that all clusters that appear in more than t percent of the input trees are represented.
The clusters are represented by the network in the "softwired" sense, which means that a cluster consists of all descendants of some ancestor, if for each reticulation one branch is "switched on" and the other branches are "switched off". See for example the following network.
The red cluster (oryza, lygeum, zea, miscanthus, pennisetum) is represented by the network, since we can switch the top branch of the reticulation off (the dashed blue branch) and the bottom branch on (the solid blue branch). This makes sense, because each cluster comes from a tree, which describes the evolution of a part of the genome. Recall that a reticulation indicates that some part of the genome evolved using the top branch and the other part of the genome evolved using the bottom branch. Hence, if the network above is the correct network, then it would not be unlikely to find the red cluster in some input tree.
See this paper for mathematical definitions and experimental results.