All the minimal sub-graphs containing SC are then analyzed with the following criteria:
- If only one stable class is on the sub-graph, all the SC are combined into that stable class. For example, in Fig. \ref{748245}A red and green are combined in \(t_{i+5}\), while yellow, brown and grey are combined in \(t_{i+3}\).
- If only two stable classes are in the sub-graph, denoted by \(C^*\) and \(C^{\prime}\), with \(C^*\) being a merger and \(C^{\prime}\) being on its ML list, then all the spurious classes in the sub-graph are combined with \(C^{\prime}\) (see sub-graph a in Fig. \ref{748245}A, where green would be \(C^*\), orange would be \(C^{\prime}\) and blue would be the spurious class).
- In any other case, the SC are discarded.
The changes in the spurious classes will modify the composition of the mergers. If the updated ML list has a single class, then the merger class can be seen as a match of that class (see sub-graph reduction a, b and c in Fig. \ref{748245}B). This step (merger reduction) might be applied recursively until all the ML lists can no longer be updated. Finally, stable classes throughout the recording might be temporarily merged (e.g. yellow and violet at \(t_{i+1}\)) or permanently merged after a certain block (e.g. yellow and violet after \(t_{i+8}\)). In those cases, the nodes are labeled in a way that represents that those classes are merged at those blocks (Fig. \ref{748245}B, d). The classes should be handled with care during these blocks; e.g. it would make no sense to compute a cross-correlation between them.