Crystal Structure Prediction of Drug Molecules in the Cloud: A Collaborative Blind Challenge Study

Abstract

Understanding the risk of multiple stable crystal polymorphs for a drug is a vital part of drug formulation and development. Computational crystal structure prediction (CSP) is a valuable tool for efficiently determining this polymorph risk. Improving the computational cost of these protocols could make polymorph screens accessible at an earlier stage of the drug discovery process. To address this need, OpenEye, Cadence Molecular Sciences (OE) partnered with GSK to organize a series of blind challenges to improve OE’s automated CSP protocol. The protocol evolved over six blind challenges, increasing in difficulty, where GSK provided minimal information about each molecule and OE predicted the low energy crystal structures. Beginning with any representation of the molecule (e.g., SMILES), OE predicted a list of possible crystal structures and ranked them with enthalpies or free energies from quantum chemical (QC) calculations. The protocol leveraged the Orion Cloud platform with highly parallelizable calculations and short wall clock times. OE’s blind predictions agreed with GSK’s experimental structures for five out of six molecules with an RMSD20 under 0.25 Å (0.5 Å for a monohydrate). The conformer was revealed for the final challenge, and OE was able to predict the crystal structure after adapting the protocol.