Fig. 6.From: Deep clustering of small molecules at large-scale via variational autoencoder embedding and K-means2D structure and similarity map for the examples randomly selected in the five clusters. For each cluster, the similarity scores between the reference compound and three test compounds were measured by the Tanimoto metric using the count-based ECFP (radius = 2, bit = 2048). The similarity weights were visualized by colors on the structure (similarity maps). Sub-structures that increase the similarity score were presented in green, whereas red indicates the oppositeBack to article page