each gene, computing QC metrics and performing dimensionality reduction. We also compared the runtime for Signac to the recently published ArchR package35 for equivalent analysis steps. For steps that are able to be run in parallel, we tested with 1, 2, 4 or 8 cores. For the PBMC dataset (Fig. 5a), we found a generally linear increase in runtime with the addition of more cells for most steps (Fig. 5b). Notably, ArchR requires a large amount of time to create the files needed to run an analysis, whereas Signac requires substantially less time for the object creation step. Overall, we found that Signac performs an end-to-end analysis of the PBMC dataset in slightly less time than ArchR (Fig. 5c). Repeating a similar analysis on the BICCN dataset (Fig. 5d), we found similar results, with both ArchR and Signac able to process consortium-scale datasets (Fig. 5e) and Signac again performing an end-to-end analysis of the full 734,000-cell dataset slightly faster than ArchR (Fig. 5f). These results provide a valuable benchmark resource for those planning experiments and estimate the time required to analyze single-cell chromatin datasets of different sizes with the Signac package.
