Fig. 3From: CASTELO: clustered atom subtypes aided lead optimization—a combined machine learning and molecular modeling methodCVAE pipeline. Left part of the Fig. shows the input preparation: the contact matrix of a time step t is grouped over the y-axis by drug atom subtypes (yellow cells are contacts). Each contact matrix of a time step t is paired to its temporal difference with a previous one \(t-\delta\) to form a dynamism tensor (violet/yellow cells are new/disappearing contacts, green cells are stable). Each atom subtype is selected in turn and fed to the CVAE. The representation in the latent space is used to cluster the time steps by similar behaviorsBack to article page