a, Benchmarking on a small dataset showed a startup cost associated with multiprocessing in distributed PyCoGAPS. b, In this small dataset, meanChiSq converges after a small number of iterations. meanChiSq values may differ slightly between distributed and single-threaded CoGAPS runs due to differing input matrix dimension. c, Benchmarking on a large single-cell dataset yielded these estimated runtimes, with Python slightly outperforming R.