Poster 3: Similarity in the Context of the Orphan Drug LegislationPedro Franco1, John Holliday1, Peter Willett1
|1Information School - University of Sheffield|
|An orphan drug is a medicinal product that is intended for the treatment of a rare disease that affects only a small number of patients, e.g., five in ten-thousand. According to the current European orphan drug legislation (European Regulation no. 141/2000 (1), the Community and the Member States shall not, for a period of 10 years, accept another orphan medicinal product, or accept an application to extend an existing marketing authorisation, for the same therapeutic indication, in respect of a similar medicinal product. In other words, a medicinal product can only be authorised for a particular rare disease if it is not similar to existing orphan drugs for that disease. |
This research project intends to establish new methodologies for the assessment of structural similarity between active substances in medicines. In this context, the main objective is to establish a correlation between the Tanimoto similarity coefficients (using various types of descriptor) obtained by computing tools and the similarity judgments of various experts in the quality and regulation of medicines from Europe, America and Asia. The experts were presented with a pair of molecules and asked to decide whether they were, or were not, similar (hence mirroring what regulatory authorities have to do when facing an application for the registration of an orphan drug). A total of 143 experts, principally from Europe but also from the USA, Japan and Taiwan, carried out this task on a set of 100 pairs of molecules chosen from Drug Bank 3.0 (2), (3), and the majority vote of the experts used to decide whether or not each pair of molecules should be considered as similar or dissimilar.
The similarity for each of the 100 pairs of molecules was also calculated using a range of 2D fingerprints (ECFP4, ECFC4, Daylight, Unity, BCI and MDL) and 1D molecular property descriptors. This was effected by means of a Pipeline Pilot protocol in which an Excel reader was used to read an Excel file containing the SMILES for each active substance as extracted from the Smiles_String_canonical fields of the DrugCards contained in Drug Bank. This enabled the establishment of a relationship between the computed similarities and the expert judgements of the form “X% of the similar pairs have a computed similarity of Y using similarity measure Z”. ROC curves were computed for each of the 2D fingerprints and the 1D molecular properties, and the performance of each similarity measure evaluated using the resulting AUC (the area under the ROC curve) value.
Our initial experiments suggest that the best results (i.e., the highest level of agreement between computed and expert judgements of similarity) are obtained using the BCI fingerprints from Digital Chemistry Limited. Future work will investigate the effectiveness of other types of similarity measure, such as pharmacophore fingerprints and ROCS shape measures, and the use of data fusion to combine different sorts of similarity measure. Having established that it is possible to obtain meaningful relationships between the computed and human judgements, our long-term aim is to provide regulatory authorities with a tool to assist them in deciding whether to approve an application for orphan drug status.
(1) Regulation (EC) No. 141/2000 of the European Parliament and Council of 16 December 1999.
(2) Drug Bank, Drug Bank 3.0, 2011, available online at http://www.drugbank.ca/ [Accessed 29 January 2012].
(3) C Knox, V. Law, T. Jewison, P. Liu, S. Ly, A. Frolkis, A. Pon, K. Branco, C. Mak, V. Neveu, Y. Djoumbou, R. Eisner, A. C. Guo, D. S. Wishart, Nucleic Acids Research, 2011, 39, 1035-1041.