Validation of pharmacophore models

The developed pharmacophore model was validated to determine its capability in differentiating the actives from the less actives and inactives and to perform virtual screening of databases. For the validation purpose, two different methods, test set prediction and decoy test prediction were employed. In the test set predication of PTP1B, a set of 19 molecules were used (Fig 1).[32,33,48] The test molecules were classified into the active (IC₅₀ < 25 μM or inhibition > 50%), less active (IC₅₀ > 25 μM or inhibition < 50%) or inactive (No activity) groups based on their known inhibitory activity. The first group of active molecules consisted of Amentoflavone, Cichoric acid, Chlorogenic acid, Molecule—3, 4, 5, 8, 14, 15, and 16. The second group consisted of Molecule—1, 2, 6, 7, and 13 while the third group consisted of Molecule—9, 10, 11 and 12 (Fig 1). The molecules were sketched and minimized in Catalyst. For each molecule a maximum of 255 conformers (with an energy threshold of 4 kcal/mol) were generated and considered for model validation. Flexible method was employed for fitting the molecule to the pharmacophore hypothesis. This method ensures an exhaustive conformational mapping even for most complex molecules. Default values were used for all other parameters in the conformational analysis. All the test set molecules were mapped on the developed pharmacophore hypothesis and “FitValue” was predicted for each test set molecule. Further, for evaluation purpose, test set molecules were divided into active, less active and inactive based on the predicted FitValues. The hit rate of pharmacophore hypothesis was calculated to analyse the efficiency of developed pharmacophore hypothesis in differentiating between actives and less actives.

^a The percentage inhibition was measured under the concentration of 100 μM. ^b The percentage inhibition was measured under the concentration of 10 μM.

For further validation of the generated hypothesis, a decoy set was generated using DecoyFinder1.1.[49] Decoys are molecules that are supposed to be inactive against a target and are used to validate the performance of the virtual screening workflow. Decoys are supposed to be inactive against a target and are used to validate the performance of the virtual screening. They were selected based on the similarity of the molecules with active ligands which is calculated considering physical descriptors such as molecular weight, number of rotational bonds, hydrogen bond donor count, hydrogen bond acceptor count and the octanol–water partition coefficient. However, they do not possess any chemical descriptors representing the active ligands.[49] For each of the 12 active PTP1B inhibitors 36 decoys were generated. Total of 432 decoy molecules and 12 active inhibitors were used to calculate various statistical parameters such as accuracy, precision, sensitivity, specificity, goodness of hit score (GH), and enrichment factor (E value). Out of these, GH and E-value are the two major parameters, playing a significant role in identifying capability of the generated pharmacophore hypothesis.^[50^]

Very few molecules are known to be the allosteric inhibitors of TCPTP. Hence, only one molecule, 16, was considered in the test set for TCPTP. Molecule 16 is a PTP1B allosteric inhibitor which has shown inhibitory activity against TCPTP (IC₅₀−40 μM, Fig 1).[48] Decoy test validation was not performed for the TCPTP pharmacophore because of the unavailability of the large number of active molecules.