Figure 2 from manuscript Sparsely-Connected Autoencoder (SCA) for single cell RNAseq data mining

Dataset used to generate Figure 2 (/fig2.png). Fig. 2: SCA analysis using a TF-based latent space. A) five clusters were detected analysing setA with griph using log10 transformed counts table. Each cluster is made by more than 90% by one cell type. A little amount of HSC is contaminating B cells, monocytes and naïve T cells (/setA/results/setA). Latent space clustering was done with SIMLR (/setA/results/setATF_SIMLR) B) QCC violin plot (/setA/Results/setATF_SIMLR/5/setA_stabilityPlot.pdf). The metric is an extension of CSS and it measures the ability of latent space to keep aggregated cells belonging to predefined clusters, i.e. those in panel A. C) QCM violin plot (/setA/Results/setATF_SIMLR/5/setA_stabilityPlotUNBIAS.pdf), this metric is also an extension of CSS and it measures the ability of the neural network to generate consistent data over multiple SCA runs. Dashed red line indicates the defined threshold to consider the latent space information suitable to support cells’ clusters. Input counts table for SCA analysis is log10 transformed.