Author's personal copy M-M. Liu er aL/European Jourmal of Medicinal Chemistry 52(2012)33-43 Considering the more druggable properties of the three flavonoids OH (compounds 5, 7 and 8)in the training set, 'principal values of 2 and 'Max-Omit-Feat values of o were assigned to these compounds, while principal'andMax-Omit-Feat values were set 1 for the other 5 compounds. Ten hypotheses(Hypos)were generated and scored as shown in Table 1. Considering it could both attain the highest score of 70.367 and accept all compounds in the Aryldiketoacid(DKA) training set statistically well enough, the hypo l was selected and was used for further validation. On the other hand. other estab- Fig 1. The structures of Aryldiketoacid(aryl-DKA)and its bioisostere galangin(mimic lished hypotheses mapped poorly to at least one component of substructure was shown in thick lines). training set(fit value <1). The Hypo 1 has four features, namely one hydrophobic feature(H), one H-bond donor(D) and two H-bond anti-HCV activities has been refined based on ligand-based and scepters(Al and a2)(Fig 4A)and compound 5 was mapped onto structure-based approaches. Hypo 1 with highest fit values of 3.99(Fig 4B). Hypo 1 was further validated by the goodness of hit(GH)scorir method [ 19, 20]. An external database of decoy set, which was use 2. Results and discussion for pharmacophore validation, was made up of other 40 indepen dent active and 1000 inactive compounds. 38 Positive compounds 2. 1. Establishment and validation of the ligand-based were successfully identified among total 55 hits and a set of statistical parameters, such as yield of actives, ratio of actives, enrichment factor(EF)and GH scores, were presented in Table 2. Known as a powerful tool to identify novel compounds with Thus the validated Hypo 1 was qualified to conduct virtual similar biological activities, the pharmacophores could be devel- screening [21] od [17]. As shown in Fig. 3, the tra set which was used to establish the pharmacophore was composed of eight NS5B inhibitors selected from literature 9, 10, 12-16 2.2. Virtual screening and molecular docking cording to the following criteria: 1. they should share certain structural diversity: 2. they should be the most active compounds Our in-house database, in which 15,568 commercially available dentified in each series: 3. they should be visually examined to natural products with searchable 3D structure were collected, was contain similar pharmacophore components in order to ensure screened by Hypol to discover potential anti-HCV candidates. 246 their similar binding models against NS5B. Due to limited activity Compounds were initially identified and most of which scale (<2 log units) and set size of the training set, the HipHop flavonoids and flavonoids glycosides. Moreover, in order to mini- module available in Discovery Studio(DS)[18 was adaptively used. mize the number of hits and to maximize the probability of positive Training set HipHop In house database(15568 Common features model (Hypo 1) Pharmacophore based virtual screening Decoy set validation Validated Initial hits(246 Pharmacophore Docking based refinement Purchased(31) Biological activity assay Active hits(20) Fig. 2. Virtual screening flow chart.Author's personal copy anti-HCV activities has been refined based on ligand-based and structure-based approaches. 2. Results and discussion 2.1. Establishment and validation of the ligand-based pharmacophore Known as a powerful tool to identify novel compounds with similar biological activities, the pharmacophores could be devel￾oped by ligand-based method [17]. As shown in Fig. 3, the training set which was used to establish the pharmacophore was composed of eight NS5B inhibitors selected from literature [9,10,12e16] according to the following criteria: 1. they should share certain structural diversity; 2. they should be the most active compounds identified in each series; 3. they should be visually examined to contain similar pharmacophore components in order to ensure their similar binding models against NS5B. Due to limited activity scale (<2 log units) and set size of the training set, the HipHop module available in Discovery Studio (DS) [18] was adaptively used. Considering the more druggable properties of the three flavonoids (compounds 5, 7 and 8) in the training set, ‘principal’ values of 2 and ‘Max-Omit-Feat’ values of 0 were assigned to these compounds, while ‘principal’ and ‘Max-Omit-Feat’ values were set 1 for the other 5 compounds. Ten hypotheses (Hypos) were generated and scored as shown in Table 1. Considering it could both attain the highest score of 70.367 and accept all compounds in the training set statistically well enough, the Hypo 1 was selected and was used for further validation. On the other hand, other estab￾lished hypotheses mapped poorly to at least one component of training set (fit value <1). The Hypo 1 has four features, namely one hydrophobic feature (H), one H-bond donor (D), and two H-bond accepters (A1 and A2) (Fig. 4A) and Compound 5 was mapped onto Hypo 1 with highest fit values of 3.99 (Fig. 4B). Hypo 1 was further validated by the goodness of hit (GH) scoring method [19,20]. An external database of decoy set, which was used for pharmacophore validation, was made up of other 40 indepen￾dent active and 1000 inactive compounds. 38 Positive compounds were successfully identified among total 55 hits and a set of statistical parameters, such as yield of actives, ratio of actives, enrichment factor (EF) and GH scores, were presented in Table 2. Thus the validated Hypo 1 was qualified to conduct virtual screening [21]. 2.2. Virtual screening and molecular docking Our in-house database, in which 15,568 commercially available natural products with searchable 3D structure were collected, was screened by Hypo1 to discover potential anti-HCV candidates. 246 Compounds were initially identified and most of which were flavonoids and flavonoids glycosides. Moreover, in order to mini￾mize the number of hits and to maximize the probability of positive Fig. 1. The structures of Aryldiketoacid (aryl-DKA) and its bioisostere galangin (mimic substructure was shown in thick lines). Fig. 2. Virtual screening flow chart. 34 M.-M. Liu et al. / European Journal of Medicinal Chemistry 52 (2012) 33e43