Efficient Handling of Molecular Flexibility in Ab Initio Generation of Crystal Structures

Abstract

A key step in many approaches to crystal structure prediction (CSP) is the initial generation of large numbers of candidate crystal structures via the exploration of the lattice energy surface. By using a relatively simple lattice energy approximation, this global search step aims to identify, in a computationally tractable manner, a limited number of likely candidate structures for further refinement using more detailed models. This paper presents an effective and efficient approach to modeling the effects of molecular flexibility during this initial global search. Local approximate models (LAMs), constructed via quantum mechanical (QM) calculations, are used to model the conformational energy, molecular geometry, and atomic charge distributions as functions of a subset of the conformational degrees of freedom (e.g., flexible torsion angles). The effectiveness of the new algorithm is demonstrated via its application to the recently studied 5-methyl-2-[(2-nitrophenyl)amino]-3-thiophenecarbonitrile (ROY) molecule and to two molecules, β-D-glucose and 1-(4-benzoylpiperazin-1-yl)-2-(4,7-dimethoxy-1H-pyrrolo[2,3-c]pyridin-3-yl)ethane-1,2-dione, a Bristol Myers Squibb molecule referenced as BMS-488043. All three molecules present significant challenges due to their high degree of flexibility.

Repulsion–dispersion parameters for the modelling of organic molecular crystals containing N, O, S and Cl

“…In the CSP approach developed in our group, 24 different energy models are used for the global search step in which thousands of possible structures are generated (CrystalPredictor 16,17,25,26 ) and in the renement step in which the ranking of structures is nalized (CrystalOptimizer 27,28 ). Recent developments in the model for the global search 16,17 have led to signicant improvements in the accuracy of the force elds for large exible molecules (such as molecule XXVI in the sixth blind test); in general, all experimentally-relevant structures are typically identied within 20 kJ mol À1 of the global minimum and carried through to the renement stage.…”

“…Recent developments in the model for the global search 16,17 have led to signicant improvements in the accuracy of the force elds for large exible molecules (such as molecule XXVI in the sixth blind test); in general, all experimentally-relevant structures are typically identied within 20 kJ mol À1 of the global minimum and carried through to the renement stage. 16,17 The latter requires a higher degree of accuracy in the lattice energy. The intramolecular energy is calculated using local approximate models derived from the isolated-molecule ab initio energy 27 computed at a chosen level of theory.…”

“…In this context, the models used in the sixth blind test included tailormade potentials tted to periodic ab initio data, 13 potentials derived from symmetry-adapted perturbation theory based on DFT (SAPT(DFT)) calculations, 14 and hybrid models combining isolated-molecule electronic structure calculations with empirical repulsion-dispersion potentials. [15][16][17] Further details of the different lattice energy models that have been used in CSP through the years and their performance can be found in the blind test papers [18][19][20][21][22] and review papers.…”

Computational Analysis of the Solid‐State and Solvation Properties of Carbamazepine in Relation to its Polymorphism

“…However, the prediction of how a solvent may impact particulate properties for crystal polymorphism generally requires the explicit calculation of intermolecular interactions between the involved molecules, using either classical molecular dynamics (MD) or quantum chemical calculations (QM) . MD calculations explore the full motion of molecules in solution over very short timescales, where the solution‐driven formation of solute/solute interactions can be identified .…”

Crystal structure prediction of flexible pharmaceutical-like molecules: density functional tight-binding as an intermediate optimisation method and for free energy estimation

“…22,29 Hence, cheaper but less accurate energy models must be utilised in the search. They can take the form of tailor-made force elds, 30 exp-6 atomistic intermolecular potentials combined with quantum-mechanical calculations of the conformational energy, 31,32 or transferrable empirical force elds. 33 The lattice energy evaluations used in the crystal structure generation ("search") stage are rarely accurate enough to give a reliable energy ranking, since the lattice energy differences between observed polymorphs are usually less than 5 kJ mol À1 .…”