Figure 5
Computational performance of CA and its adaptations. (A) Plot comparing runtime for standard CA and glmPCA on ten datasets, selecting down to 1500 features in each. Standard CA consistently runs in under a minute, even for datasets with over 20,000 cells, while glmPCA scales less favorably and requires over an hour for the equivalent input matrix (1500 features x ~ 22,000 cells). (B) Plot comparing runtime with increasing number of features in the Aztekin Xenopus tail dataset, across the CA adaptation methods. Since they use similar routines, their runtimes are fairly similar. (C) Plot comparing runtime with increasing number of features in the Zhengmix8 dataset, across the CA adaptation methods. In both (B) and (C), it is notable that even with an order of magnitude more features, CA and its adaptations run in a fraction of the time glmPCA takes.
