Figure 3

Diabetes phenotypes identified using TDA. (a) The network resulting from TDA, where each node contains patients that have similar diabetes embeddings. (b) The distribution of known cardiovascular risk factors and pre-existing diseases within the four identified regions/subtypes. In the TDA network, nodes correspond to patients with similar representation; nodes are connected if they share at least one patient (see “Materials and methods”). The nodes and their colours represent the 4 clusters obtained using the METIS partitioning method. A fully interactive version of this figure that shows further details at the node and cluster level is available at: https://deepmedicine.github.io/TDA/multiple.html.