Skip to main content


Fig. 5 | Journal of Cheminformatics

Fig. 5

From: KekuleScope: prediction of cancer cell line sensitivity and compound potency using convolutional neural networks trained on compound images

Fig. 5

Predictions for the test set molecules. Observed against predicted pCI50 values for the test set compounds calculated using ConvNets (top panels) or RF models (bottom panels). The results for the ten repetitions are shown (in each repetition the molecules in the training, validation and test sets were different). Overall, both RF models and ConvNets generated comparable error profiles across the entire bioactivity range considered, showing Pearson’s correlation coefficient values in the 0.72–0.84 range

Back to article page