Assessment and Improvement of Structure Generation in Crystal Structure Prediction Protocols

Abstract

Crystal structure prediction (CSP) is a challenging task in physical chemistry, involving the prediction of 3D crystal structures from single-molecule structures without experimental input. CSP has significant applications in industries such as pharmaceuticals, organic electronics, and dyes. Force field (FF) methods are commonly used in the initial stage of CSP due to their computational efficiency. A complete CSP protocol was completed, and a benchmark analysis of accessible FFs was conducted using an evolutionary algorithm structure generator. The performance of the FFs was assessed based on their ability to identify the experimental polymorphs of compounds in the PV17 benchmark set, plus 5-fluorouracil. The generalized AMBER force field (GAFF) was identified as the optimal choice for CSP at this time, as it demonstrated a high rate of polymorph match identification with low relative energies. The results from this benchmark provide insights into key FF features necessary for successful CSP protocols.