2.3. Evaluating the accuracy of inferred clones and phylogenies

To evaluate the accuracy of inferred clone phylogenies, we computed the RF distance using the phangorn R package [26]. RF distance is a standard approach to evaluate the similarity of two phylogenies in molecular evolution and phylogenetics. We converted simulated cell phylogenies to clone phylogenies, where clones were groups of cells with no sequence differences. We paired each simulated (true) clone to an inferred clone that contained the largest number of cells that belonged to the simulated clone. We allowed a single inferred clone to be paired with more than one simulated clone. When more than one simulated clone was paired, the tip of the inferred phylogeny was duplicated because the RF computation required the same number of tips in the phylogenies. Also, we pruned inferred and true clones that were not paired. For example, a simulated clone did not have an inferred clone pair when those simulated cells were excluded in the inferred clone annotation, e.g., false-positive detection of doublets. We counted the number of total partitions and those not found in the other phylogeny for each phylogeny. The number of pruned simulated clones for the RF computation was counted as the number of partitions not found in the inferred phylogeny. Overall, RF is not designed to be used for phylogenies with different numbers of tips, and errors of missing and additional clones cannot be accurately assessed using RF distances. Therefore, to further assess errors in the size of inferred phylogeny (the number of tips), we also reported the difference in the number of clones inferred and the true clone count.