General Computational Algorithms for Ab Initio Crystal Structure Prediction for Organic Molecules

Abstract

The prediction of the possible crystal structure(s) of organic molecules is an important activity for the pharmaceutical and agrochemical industries, among others, due to the prevalence of crystalline products. This chapter considers the general requirements that crystal structure prediction (CSP) methodologies need to fulfil in order to be able to achieve reliable predictions over a wide range of organic systems. It also reviews the current status of a multistage CSP methodology that has recently proved successful for a number of systems of practical interest. Emphasis is placed on recent developments that allow a reconciliation of conflicting needs for, on the one hand, accurate evaluation of the energy of a proposed crystal structure and on the other hand, comprehensive search of the energy landscape for the reliable identification of all low-energy minima. Finally, based on the experience gained from this work, current limitations and opportunities for further research in this area are identified. We also consider issues relating to the use of empirical models derived from experimental data in conjunction with ab initio CSP.

Prediction of the crystal structures of axitinib, a polymorphic pharmaceutical molecule

“…1) and often arises from the presence of rotatable bonds (Cruz-Cabeza and Bernstein, 2014) with a deformation energy of the same order of magnitude as the energy change on packing (Kazantsev et al, 2011a). It is thus essential to account accurately for molecular flexibility during the course of energy minimization; a discussion of the pitfalls of neglecting or overly restricting flexibility can be found in Pantelides et al (2014). The need to explore conformational variation greatly increases the complexity of the optimization problem and has been a key driver for recent theoretical and algorithmic developments of our CSP framework.…”

“…In addition, it is necessary to be able to predict the crystal structures of salts, co-crystals and solvates as such systems are frequently used to enhance product effectiveness or to facilitate manufacturing. In a recent publication (Pantelides et al, 2014), we set out the design requirements that systematic CSP methodologies should meet in order to find wide applicability in practice, such as (i) a high degree of automation, with limited dependence on user insight; (ii) a consistent and general physical basis; (iii) a high degree of reliability; (iv) a high degree of accuracy, but with reasonable computational cost. This last requirement is particularly challenging: experience in crystal structure prediction has shown that it is essential to carry out an exhaustive search of the energy landscape covering millions of potential structures, and that the relative energies of computed crystal structures are highly dependent on the accuracy of the energy model, with electronic structure calculations providing the most reliable results (Bardwell et al, 2011).…”

“…Inaccuracies can arise from the neglect of specific contributions to the crystal energy (e.g., polarizability and entropic effects). In fact, such contributions can be captured partly (Pantelides et al, 2014) within the empirical term of the computational model used here (nominally the "repulsion/dispersion" term) whose parameters are fitted to energetic and structural data from a number of crystal structures. In this context, it is interesting to note that, while the importance of the accurate modeling of the electrostatic and intramolecular contribution has gained wide understanding and acceptance in recent years, much less attention has been given to the impact that the empirical term contribution may have on the final outcome.…”

“…Together with the papers summarizing the results of the five blind tests to date (Bardwell et al, 2011;Day et al, 2009Day et al, , 2005Lommerse et al, 2000;Motherwell et al, 2002), these provide an excellent survey of the field. In the present paper, we focus on a specific systematic approach that has been developed in our group (Pantelides et al, 2014). The algorithms on which this approach is based have been successfully used in several of the examples discussed so far (Bardwell et al, 2011;Bhardwaj et al, 2013;Ismail et al, 2013;Kazantsev et al, 2011b;Price et al, in press;Vasileiadis et al, 2012).…”

“…Several reviews have recently been published on the current state-of-the-art in CSP, covering one or more methodologies (Abramov, 2013;Atahan-Evrenk and Aspuru-Guzik, 2014;Day, 2010Day, , 2011Day, , 2012Kendrick et al, 2011;Pantelides et al, 2014;Price, 2013Price, , 2008aPrice, , 2008b). Together with the papers summarizing the results of the five blind tests to date (Bardwell et al, 2011;Day et al, 2009Day et al, , 2005Lommerse et al, 2000;Motherwell et al, 2002), these provide an excellent survey of the field.…”