Theoretical Calculations to Assist Experimental Crystal Form Screening.

APIs can crystallise in multiple distinct forms, each with their own physical properties such as dissolution rate, crystal habit or melting point. This phenomenon, known as polymorphism, has long been recognised by pharmaceutical industry and regulatory bodies alike as playing a crucial role in the formulation of drugs marketed in crystalline form. Experimental screens for polymorphs, salt forms or co-crystals are an important instrument in determining the most suitable solid form. Especially the thermodynamic stability landscape needs to be explored thoroughly, as identification of the thermodynamically most stable form is paramount. Discovery of a more stable polymorph after a plant has been commissioned has implications for patents and for the manufacturing process--especially if the more stable polymorph is discovered by a competitor. The Achilles\' heel of experimental polymorph screens is kinetics: the route to the thermodynamically most stable form may not be accessible during the experiment, causing the most stable form to be missed completely. Theoretical polymorph screens (generally referred to as crystal structure prediction) do not suffer from kinetics and are therefore the instrument of choice to complement experimental polymorph screens in order to check if possible stable polymorphs have been overlooked. In the early days, such in silico polymorph screens were highly unreliable due to the quick and dirty nature of the methods used in evaluating the relative thermodynamic stabilities of the predicted polymorphs. The availability of relatively cheap computing power, however, has changed this completely, as demonstrated in the 2007 Crystal Structure Prediction Blind Test when the application of high-quality quantum-mechanical calculations correctly predicted the crystal structures and relative stabilities of all four target compounds. The strengths and weaknesses of in silico polymorph screens with such high-quality quantum-mechanical calculations will be demonstrated using unpublished results on the antiretroviral agent Efavirenz as an example.