Skip to main content
. 2021 Sep 28;49(22):e127. doi: 10.1093/nar/gkab823

Figure 3.

Figure 3.

Simulation results for the ‘large-condition-effect’ setting. (A) Proportions of cell pairs with edges across different numbers of layers of MLG constructed from SNN graphs of PCA and cNMF, Seurat, and Liger (with 20 neighbors in SNN graphs and 10 latent factors in low-dimensional embeddings). The boxplots depict the proportions across all the simulation replicates. (B) Estimated signal-to-noise ratios of SNN graphs constructed from different low-dimensional embeddings and their multilayer graph across all the simulation replicates (with 20 neighbors in SNNs and 10 latent factors in low-dimensional embeddings). (C) Louvain clustering accuracy of SNN graphs as a function of numbers of neighbors in SNN graph construction (with 10 latent factors in low-dimensional embeddings). (D) Adjusted Rand index comparison of Louvain and k-means clustering of SNN graphs from different low-dimensional embeddings and their MLG as a function of number of latent factors in the low-dimensional projections (with 20 neighbors in SNN graphs). ARI values of supervised knn classifiers for individual SNN graphs are provided as reference.