Fig. 8
From: ENGEP: advancing spatial transcriptomics with accurate unmeasured gene expression prediction
Comparison of running time and peak memory. Running time (a) and peak memory (b) of different methods. We use different methods to predict 5000 unmeasured genes in spatial dataset containing 3700 cells and 254 genes with reference datasets comprising 30,000 genes and varying numbers of cells, specifically, 10,000, 50,000, 100,000, 200,000, and 500,000 cells. We run both Tangram-cell and Tangram-cluster on our GPU platform, except for when we run Tangram-cell with a reference containing 200,000 cells. In that case, we encounter a memory error while using GPU, so we switch to using CPU instead