Based on Mantis-ML's predictions it is possible to construct a gene or disease network. The network is constructed through considering the Mantis-ML feature importances for a given disease then finding other diseases that are 'close' to them based on this vector of importances. In particular, we employ a nonlinear t-SNE embedding of all disease scores and project these down to two dimensions (shown below).

Click on a cluster to view mean feature importance.