scCODA uses HMC sampling for parameter inference. Therefore, the most important factor in runtime is the duration of one HMC sampling step. To isolate the HMC sampling process from the model initialization and post-sampling analysis steps, we applied scCODA twice to each dataset, sampling chains of length 1,000 and 2,000, respectively. We measured the execution time for both instances and divided the time difference by 1,000—the difference in chain length—to gain an estimate for the execution time per sampling iteration (Supplementary Fig. 12). All operations were executed on an Intel(R) Xeon(R) Gold 6126 processor. The memory consumption of a single run of scCODA in default settings should not exceed 2 GB.
