In silico profiling nanoparticles: predictive nanomodeling using universal nanodescriptors and various machine learning approaches

Abstract

Rational nanomaterial design is urgently demanded for new nanomaterial development with desired properties. However, computational nanomaterial modeling and virtual nanomaterial screening are not applicable for this purpose due to the complexity of nanomaterial structures. To address this challenge, a new computational workflow is established in this study to virtually profile nanoparticles by (1) constructing a structurally diverse virtual gold nanoparticle (GNP) library and (2) developing novel universal nanodescriptors. The emphasis of this study is the second task by developing geometrical nanodescriptors that are suitable for the quantitative modeling of GNPs and virtual screening purposes. The feasibility, rigor and applicability of this novel computational method are validated by testing seven GNP datasets consisting of 191 unique GNPs of various nano-bioactivities and physicochemical properties. The high predictability of the developed GNP models suggests that this workflow can be used as a universal tool for nanomaterial profiling and rational nanomaterial design.