Fig. 3From: KekuleScope: prediction of cancer cell line sensitivity and compound potency using convolutional neural networks trained on compound imagesAnalysis of the residuals. Residuals for the ConvNets (top panels) and RF (bottom panels) models for the cytotoxicity data sets. Overall, the residuals for both types of models show comparable variance across the bioactivity range (a) and are centered around zero (b), indicating that compound images permit to model the activity of small molecules across a dynamic range of pIC50 valuesBack to article page