Fig. 5
From: Extended study on atomic featurization in graph neural networks for molecular property prediction
Feature importance scores for four predictions of a model trained with representation F-N on ESOL (scaffold). Importance score value (on y-axis) for each feature (on x-axis) is represented with a bar whose color denotes feature group and intensity denotes feature value – features with non-zero value for at least one atom in a molecule are shown in vibrant colors while features which are equal to zero for all atoms are shown in pale colors. A correlation between feature values and importance scores can be observed