Clustering of Small Molecules

Clustering of small molecules finds a diversity of applications in chemistry and, in particular, in the fields of cheminformatics and drug discovery. It may be used directly as an unsupervised machine-learning strategy to identify existing patterns in a chemical data set or libraries or integrated into supervised machine-learning studies to partition a sample of compounds into representative subsamples (e.g., training and validation data). It may also be applied to select which in silico hits from a virtual screening campaign will be submitted to experimental confirmation, or to define which hits emerging from a “wet” screening campaign will be prioritized for further development or characterization. Here, we review general strategies to validate the output of a clustering algorithm and discuss current challenges and possible future directions in the field of small molecule clustering.