Skip to main content
Fig. 2 | Genome Medicine

Fig. 2

From: Diagnostic Evidence GAuge of Single cells (DEGAS): a flexible deep transfer learning framework for prioritizing cells in relation to disease

Fig. 2

Simulation study and baseline comparisons of DEGAS framework. A 5000 simulated cells from Splatter with 4 cell types where one of the cell types has two subtypes. Cell type 4 is composed of two subtypes that are specific to either disease or normal patients. In total, 2000 of these cells were used to generate the 600 simulated patients in B–D and 3000 were used as the cell input to our DEGAS models. E Optimal cluster number (4 clusters) based on average silhouette width for the 3000 cells not used to generate patients. F The same 3000 cells used as the cellular input colored by their cluster. G DEGAS comparison to Augur in simulation 1. H DEGAS comparison with Augur in simulation 2. I DEGAS comparison with Augur in simulation 3. J–L DEGAS-calculated disease association from each simulation overlaid onto 3000 cells. The violin plot in the bottom left corner is deconvolution cell type proportion for cell type 1 in simulation 1 patients (J), cell type 4 proportion in simulation 2 patients (K), and cell type 4 proportion in simulation 3 patients (L)

Back to article page