Fig. 4.

Comparison of predicted probabilities from top 3 CANTARE models. Predictions from 3 logistic regressions, each seeded with the network neighborhood of a different microbe. Predictors in Model S are a proper subset of the predictors in Model A. The predictions from these models are more similar to each other (r = 0.948) than they are to the E model (r = 0.739 and r = 0.719), with which they have no overlapping predictors. Black lines in the cells on the diagonal represent the distributions of the predictions