Fig 3. Mean performance metrics for random forest binary classification models trained on data sets containing imputed IC₅₀ values.

IC₅₀ values were imputed by the k-nearest neighbors (k = 9) algorithm (left) and the logistic regression with lasso (‘lassoC’) algorithm (right). Percentages of imputed values range from 10% to 40%. An IC₅₀ activity cutoff of 1 μM was applied. The kappa statistic (κ) has been multiplied by 100 for scaling.