Examples

ZOMBI2 is a tool to generate datasets in which the user knows everything about them. Let's see how that is used in practice. Each example asks a question about a real method and answers it with a dataset where the truth is known. Every file behind each example, the scripts, the results and the exact reproduction recipe, is in the repository under analyses/.

Can RED be trusted? GTDB uses Relative Evolutionary Divergence to align taxonomic ranks across the tree of life; the measure assumes branch length tracks time. We measure how rate-variable real archaea are, simulate trees with the same variability, and test RED on trees whose node ages are known.

Can BiSSE find the gene that drives speciation? A gene family raises the speciation rate of the lineages that carry it, in a joint run where the genome shapes its own tree. The dataset includes a matched null and an undriven family, so the standard test can be scored on known truth: how often it fires when nothing drives, and how often it finds the gene that does.

Which rearrangement parameters can be recovered from gene order? Genomes are simulated at known inversion and translocation rates and the parameters inferred back. The rates are recovered, including the mix of the two event types. The size of the events is not, and fixing it at a wrong value biases the rates that are recoverable.

Can Pagel's test detect a feedback? A habitat drives gene loss across the genome, and the absence of one gene family drives the switch rate into the parasitic habitat, a feedback closed in one joint run. Pagel's test detects the feedback and the switch-rate connection in about nine replicates of ten, largely misses the loss connection, and rejects at the nominal rate with no connections and on a control family carried on the same trees.

Who trades genes with whom? Gene transfers are drawn preferentially between lineages sharing a habitat, and the habitats of ancestral lineages are read back from the transfer network that ALE infers. A vote over the inferred transfers classifies all ancestral branches at 92% accuracy, equal to the ceiling a perfect reconciliation could reach; the cost of extinction, of dead donors and of the reconciliation itself is measured at every step.