Information gleaned from crystal structure databases has previously been reported on several pharmaceutically relevant compounds to make knowledge-based predictions of polymorphism. Access to a large dataset that is highly relevant to the molecules under study is considered to be essential for these studies. We present a survey of the GlaxoSmithKline (GSK) database of small molecule crystal structures containing X-ray diffraction results from GSK and heritage companies from the past 40 years for this purpose. These structures were collected at different stages of the pharmaceutical pipeline and are not limited to marketed products. We found that the GSK database matches the CSD Drug Subset in terms of crystal descriptors, but not in the diversity of solid form space. Applying the hydrogen bond propensity model to GSK polymorphs has demonstrated the increased value in using combined published and proprietary data sources to build the training data sets. Within GSK, we have also shown the value of applying knowledge-based predictions in the de-risking of active pharmaceutical ingredient forms of development candidates. The work described here illustrates the importance of database curation to improve the accuracy of the results obtained.
