Adipose HOMA-IR master regulator analysis.

(A) Cross-validated prediction performance of random forest regression models with the number of predictors sequentially reduced by five. Models were trained to predict ln(HOMA-IR) from regulator activities 12 times, each with a unique seed. The plot compares the number of predictors included in the model versus the mean cross-validation error and error bars indicate the standard deviation of the 12 analyses. (B) Rank order for the top 100 regulators by importance to ln(HOMA-IR) prediction by the final random forest model. The importance is measured as the percent increase in the mean squared error (MSE) upon permutation of the regulator across all trees of the random forest.

(PDF)