Molecular shape as a (useful) bias in chemistry

Abstract

One of the molecular properties most intuitive to the human perception is the geometrical shape. However, when exploring a large chemical space the determination of shape needs to be automated. We present a fast and simple approach to identify a molecule as linear, planar, cube, cuboid, disk, elliptical disk, spheroid and sphere which is more fine grained than existing approaches. The method is applied to more than one billion molecules ranging from small organic molecules to whole proteins. The results show that current chemistry research is biased towards planar geometries. Moreover, we demonstrate that our molecular shape classification correlates with sought-after properties like the band gap, dipole moment, and heat capacity. This allows to increase the efficiency of molecular design studies by driving high-throughput-screening efforts towards desired values of molecular properties.