Skip to main content
. 2018 Dec 18;16(5):320–331. doi: 10.1016/j.gpb.2018.08.003

Figure 4.

Figure 4

Analysis of the model stability and parameter settings of VASC

A. The iteration process using the Pollen dataset [27]. The change of loss values of the whole network as shown in Equation (2) versus iteration epochs is shown on the left and the right part is the change of NMI and ARI values versus iteration epochs is shown on the right. B. The iteration process using the Biase dataset [18]. C. The stability of VASC. The boxplots were generated based on 20 repeated runs with (green) or without (orange) the ZI layer. Tests were performed on the Biase [18], Goolam [22], Pollen [27], and Yan [30] datasets. D. The down-sampling test on cell numbers based on the Pollen dataset [27]. VASC was run on 10%–100% randomly-sampled cells of the original dataset. E. The down-sampling test on read numbers based on the Pollen dataset [27]. F. The effects of the dimensions (ranging from 2 to 30) for the latent variables based on the Pollen dataset [27]. ZI, zero-inflated.