Close
  Indian J Med Microbiol
 

Figure 5: Visualizing networks of spatial interactions from pointwise mutual information maps. The pointwise mutual information map in the middle denotes the relative probability of finding two co-occurring phenotypes i and j in reference to a background distribution. In the colorbar above the pointwise mutual information map, red/blue indicates highest/lowest possible co-occurrence, and black indicates an absence of interactions. The stem plots to the right describe the eight phenotypes learned from the data, where each stem plot represents the relative estrogen receptor, human epidermal growth factor 2, and progesterone receptor intensities of the phenotype (left to right). The labels for each stem plot (1-8) correspond to the rows and columns of the pointwise mutual information map. This map allows us to probe any tumor sample for networks of spatial interactions that contribute to the pointwise mutual information calculation. We display representative networks of spatial interactions for three different pointwise mutual information map entries. The two networks shown in yellow are examples where phenotype 6 spatially co-occurs with itself more frequently than expected from the background distribution. The two networks shown in green indicate two spatial networks where phenotype 5 spatially co-occurs with itself as would be expected from a random phenotyping of cells, given phenotype background probabilities. The two networks shown in blue portray interactions between cells of phenotypes 2 and 3 spatially co-occurring, which happens less than is expected from the background distribution. In each of these cases, the nodes in these graphs are the spatially co-occurring cells of a specific phenotype, and edges are only drawn to cells in spatial proximity. Depending on individual tumor graph statistics, these spatial relationships may be localized or ubiquitous throughout the tumor

Figure 5: Visualizing networks of spatial interactions from pointwise mutual information maps. The pointwise mutual information map in the middle denotes the relative probability of finding two co-occurring phenotypes <i>i </i>and <i>j </i>in reference to a background distribution. In the colorbar above the pointwise mutual information map, red/blue indicates highest/lowest possible co-occurrence, and black indicates an absence of interactions. The stem plots to the right describe the eight phenotypes learned from the data, where each stem plot represents the relative estrogen receptor, human epidermal growth factor 2, and progesterone receptor intensities of the phenotype (left to right). The labels for each stem plot (1-8) correspond to the rows and columns of the pointwise mutual information map. This map allows us to probe any tumor sample for networks of spatial interactions that contribute to the pointwise mutual information calculation. We display representative networks of spatial interactions for three different pointwise mutual information map entries. The two networks shown in yellow are examples where phenotype 6 spatially co-occurs with itself more frequently than expected from the background distribution. The two networks shown in green indicate two spatial networks where phenotype 5 spatially co-occurs with itself as would be expected from a random phenotyping of cells, given phenotype background probabilities. The two networks shown in blue portray interactions between cells of phenotypes 2 and 3 spatially co-occurring, which happens less than is expected from the background distribution. In each of these cases, the nodes in these graphs are the spatially co-occurring cells of a specific phenotype, and edges are only drawn to cells in spatial proximity. Depending on individual tumor graph statistics, these spatial relationships may be localized or ubiquitous throughout the tumor