Saelens, W. et al. Nat. Biotechnol. 37, 547–554 (2019).

Single-cell transcriptomics data contain a wealth of information, not merely on the composition of a sample at the time it was taken, but also on the dynamic processes that gave rise to it. Over 70 computational methods have been developed that order cells along a pseudotime trajectory on the basis of similarities in their expression patterns. This myriad of tools can make the choice of a suitable one difficult. Saelens et al. have benchmarked 45 tools on 110 real and 229 synthetic datasets according to their accuracy, scalability of cells and features, stability after subsampling, and usability. Each method is also characterized by the type of trajectories it can infer, from a simple linear path to multifurcations and tree structures to more complex connected or disconnected graphs. The researchers evaluated whether the combination of any two methods yielded a model with a higher score. Their conclusion is that the choice of tool should be driven by the known or expected trajectories of the data. A practical user guide (https://benchmark.dynverse.org) will be helpful in guiding such choices.