Choose "algorithms" and "level-k network consensus".

A popup box will appear, which gives you several options.

If you do **not** check the box "construct only one network", the program will search for multiple solutions (it is not guaranteed to find all solutions). If you do check the box, the program might be faster because it will search for just one solution.

If you do **not **check the box "construct only networks that display the trees", the program will output networks that display all clusters of the input trees. If you do check the box, the program will output networks that display the trees themselves (and hence also their clusters). Note that the latter option might need more reticulations, and possibly a longer running time. If the input consists of two trees then both options will give the same number of reticulations (but not necessarily the same networks).

The final option allows you to filter out badly supported clusters. You can enter a threshold percentage *t*. If you enter 0, all clusters from all input trees are used. If the computations take too long, or if an overly complicated network is constructed, then you can decide to enter a higher threshold percentage *t*. The program will in that case only use clusters that appear in more than *t* percent of the input trees. It is also recommended to preprocess the trees by collapsing poorly supported branches. This is not yet implemented in Dendroscope.

After clicking "OK". The program starts searching for networks. Do not worry if the progress bar doesn't move - it is not very accurate. To get a better indication of the algorithm's progress, click "window" and "message window" (preferably before starting the computations). You will then see which level the algorithm is at. Note that the algorithm is fast even for large number of leaves as long as the level is low. Usually, it will reach level-7 or 8.

For example, for this input file, you get the following network if you **do not** check the box "construct only networks that display the trees". This network displays all clusters from the input trees and has one reticulation.

If you **do** check the box "construct only networks that display the trees", you get the following three network, each of which displays all input trees and has two reticulations. Hence, for this data set, more reticulations are needed to display the trees than to display the clusters from the trees.

For this grass data file, CASS constructs the network below.

If you let Cass search for more solutions, it finds 2619 different networks that all display the input trees.

If the computations take too long, you can choose to construct a "galled network consensus" or a "cluster network consensus" instead. These methods might use more reticulations but are more efficient.