Fig. 3
From: A rank-based marker selection method for high throughput scRNA-seq data
Computational resources used by the marker selection methods. In both figures, the data set size is the number of entries in the data matrix X: it is given by n×p, the number of cells times the number of genes. The data sets that we consider in this work are indicated in the figures. The total computation time required to select markers on one fold (CPU time, calculated as number of processors used multiplied by the time taken for marker selection) is shown in (a); the total memory required during these trials is shown in (b). Elastic nets scales poorly in (a), so it is only run on PAUL and ZEISEL. Both edgeR and MAST are limited by memory on ZHENGFILT (see (b)); this prevents their application to the larger data sets. scVI also requires a GPU while it is running; this prevents us from testing it on the larger data sets. RANKCORR, the t-test, Wilcoxon, and logistic regression all use 8 GB to run on ZHENGFULL and 80 GB to run on 10XMOUSE. See the marker selection methods for more details