Chemically similar drugs often bind to biologically diverse targets, making it difficult to predict what off-target effects a drug might have by protein structure or sequence alone. A new informatics tool called Similarity Ensemble Approach (SEA) uses a different strategy; it groups receptors according to the chemical similarity of their ligands and can identify unknown relationships between ligands and receptors that can be tested experimentally.
"The ability to identify off-target effects in silico before further experimental verification would be invaluable."
SEA, described in detail in Nature Biotechnology, is the result of a collaborative effort using computational expertise from Brian Shoichet's laboratory combined with experimental confirmation carried out in Bryan Roth's laboratory. The groups took data from the Elsevier MDL Drug Data Report (MDDR) comprising 65,000 ligands annotated for 246 specific drug targets and used this to compare the chemical similarity between ligand sets. This allowed them to map targets according to the chemical similarity of their ligands.
Although no biological data were used, biologically related targets were found to be clustered together. The chemical similarity comparison also identified groups of receptors that were unrelated by sequence and structure but had ligands with common chemical features. Moreover, within these clusters, it was also possible to identify individual receptor subtypes, something that could be invaluable to medicinal chemists who are designing subtype-selective drugs.
Having made meaningful comparisons of chemical similarity, the authors looked for ligands that were predicted to have off-target activity to see whether their predictions could be confirmed experimentally. An interesting new property was observed for methadone, which is known to act at NMDA (N-methyl-D-aspartate) and 
-opioid receptors. In this study, methadone and its analogues seemed to be more similar to the antimuscarinics ligand class, a prediction that was confirmed by cell-based functional assays in which methadone was shown to be a 1-M antagonist of M3-receptor activity.
A screen of PubChem compounds against the 246 MDDR targets found more off-target effects. Many of these could be accounted for by differences in annotation term rather than interesting pharmacology. However, about 30 compounds had good 'expectation values' for binding to biologically unrelated MDDR targets, and two of these — emetine and loperamide — were tested further experimentally.
As predicted by SEA, emetine and loperamide acted as competitive inhibitors of the 
2 adrenoceptor and neurokinin NK2 receptor, respectively, revealing unknown pharmacology for both drugs. Emetine's antagonistic effects on the 2 adrenoceptor explains some of the drug's side effects, and although a link between loperamide and the neurokinin pathway via opioid-receptor signalling has been postulated, this is the first evidence of a direct effect of the drug on neurokinin receptors.
Such experimental validation suggests that other predicted relationships might warrant investigation, something that should be encouraged by the free availability of the linkage maps online. Certainly, the ability to identify off-target effects in silico before further experimental verification would be invaluable and could potentially reduce the requirement for costly and time-consuming functional assays. But the authors also believe it might be useful in one of industry's favourite strategies — drug repositioning — by facilitating the search for novel targets of marketed drugs.