Figure 2.

Models of chemical features accurately predict inhibitors of SARS-CoV-2 targets.a) Pipeline for fitting and validating models that predict IC₅₀, K_i, or AC₅₀ or a classification score, which reflects broad inhibitory activity against the listed viral targets. b) Left, mean absolute error (MAE) in predicting the log transformed endpoints (IC50, Ki, AC50). Right, classification of chemicals for broad inhibition or activity against targets, validating using the area under the receiver operating characteristic (ROC) curve (AUC). Plots are for 10-fold cross validation, repeated 5 times. The model predictions are from an ensemble of three support vector machines (SVM), trained on different chemical feature sets or in some cases SVM and Random Forest. c) Left, external test set performance for regression models, where possible. Right, external test set performance for classification models, where possible. More comprehensive performance data in Supplementary Information 1.