Statistical cluster analysis of pharmaceutical solvents

High efficiency in polymorph screening and crystallization optimization can be gained by judicious selection of solvents for the study design. Examination of all 57 (classes 2 and 3) pharmaceutical solvents may enable a complete study design but is costly in terms of time and resources. Based on a 17 descriptor dataset specifically constructed for all the classes 2 and 3 pharmaceutical solvents recognized by the International Conference of Harmonization (ICH), an optimal two-stage cluster analysis was carried out together with principal component analysis as a dimensionality and colinearity reduction pre-processor. Both hierarchical average linkage cluster analysis and non-hierarchical K-means cluster analysis converged on a 20-cluster solution with strong statistical criteria support and excellent agreement in cluster memberships, which can be reasonably interpreted from a chemical perspective. This 20-cluster solution is offered as an option for design of more efficient solid state screening studies. Rather than designing a polymorph screen to include all 57 solvents, the inclusion of a single member from each of the 20 clusters would be expected to adequately represent the full range of solvent properties exhibited by the entire 57 member solvent set.