>home>Papers>Keynote Speakers

The structure of the crystals formed by an organic molecule often has a significant effect on properties such as density, colour, solubility, rate of dissolution and melting point. Consequently, it is of major interest to many sectors of the process industries (e.g. pharmaceuticals, pigments) as it influences both the product characteristics (e.g. the bioavailability of a drug or the colour of a pigment) and the ease of processing during production. Many organic molecules exhibit polymorphism, i.e. they form two or more crystal structures which differ only slightly in free energy and can even coexist under the same thermodynamic conditions. Polymorphism is of great practical importance as properties can vary significantly between polymorphs of the same molecule. A well-publicized example of the adverse effects of the existence of unknown polymorphs is the case of Ritonavir (NorvirTM) that was launched by Abbott Laboratories in 1996; two years later, a more stable form with significantly different dissolution properties was discovered, causing expensive delays in bringing the new drug to the market. This paper presents a comprehensive methodology for the prediction of crystal structures using only the atomic connectivity of the molecule under consideration. The method developed is based on the global minimisation of the molar enthalpy of the crystal. The modelling of the electrostatic interactions is accomplished through the use of a set of distributed charges that are optimally selected and positioned based on results of quantum mechanical calculations. A two-phase stochastic/deterministic global optimisation method is used for the identification of the local minima of the lattice enthalpy surface. The method uses low discrepancy sequences to generate a number of initial guesses in the space of the optimization variables. It then initiates local optimisation calculations with a sequential quadratic programming algorithm. A parallelized implementation of the algorithm allows minimisations from many thousands of initial guesses to be carried out in reasonable time. Both rigid and flexible molecules are handled. In the latter case, intramolecular energy is computed as a function of key internal degrees of freedom (e.g. torsion angles) by fitting a Hermite interpolant on values computed using quantum mechanical calculations. The algorithm has been applied successfully to the prediction of the crystal structures of a large set of compounds.