Skip to main content
Fig. 3 | Genome Medicine

Fig. 3

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

Fig. 3

DEGAS validation in GBM and AD. DEGAS output of the distribution of GBM subtypes in single cells from five GBM tumors. Four of the five tumors had known GBM subtype information from Patel et al. (MGH26: Proneural, MGH28: Mesenchymal, MGH29: Mesenchymal, and MGH30: Classical, indicated by red boxes) which were recapitulated by DEGAS. The subtype information for the tumors, MGH26, MGH28, MGH29, and MGH30 were derived from Patel et al. where MGH31 did not have a clearly defined GBM subtype. The association of cells assigned to each subtype were plotted for each tumor; A MGH26, B MGH28, C MGH29, D MGH30, and E MGH31. Median values are marked by a diamond in each of the violin plots. F The death association centered around 0 is overlaid on all of the single cells from the five tumors (indicated by color). G DEGAS output of AD association for each single cell. The AD association score is indicated by the color and is overlaid onto AIBS single cells. This plot shows the negative AD association in neuron cells and positive AD association in Microglia. H–I There also appeared to be a subpopulation of astrocytes with positive AD association. The astrocytes were plotted separately and colored by AIBS Astrocyte subtypes (H) and GFAP expression, a disease-associated astrocyte marker (I). J Comparison of DEGAS-derived AD associations for single cells from AD and Normal control samples from Grubman et al. K–M Targeted analysis of microglia from Grubman et al. including the AD associations overlaid onto microglia (K), AD association comparing AD status of patient sample from which the cells were sampled (L), and PCC between AD association with HAM marker genes comparing up- and downregulated HAM marker genes (M). Significance values: n.s. (not significant), • (0.1), * (0.05), ** (0.01), *** (0.001)

Back to article page