First, a graph is created for the whole recording, where nodes that could be mergers, new classes, or matches, are connected with edges across consecutive blocks according to the tracking criteria explained above (see Fig. \ref{748245}A). Next, all the graph components (isolated sub-graphs) are obtained; in the case of Fig. \ref{748245}A , 3 sub-graphs a-bc-d, and e. If no "stable classes" are included on a given sub-graph, all the classes in that sub-graph are discarded (see black node in sub-graph e in Fig. \ref{748245}A). Then, minimal sub-graphs containing spurious classes are further isolated by removing edges linking nodes from the same stable classes (i.e. edges connecting nodes of the same color). For example, in Fig. \ref{748245}A , these sub-graphs can be found within the sub-graphs ab and c, with resulting sub-graph d having no spurious classes.