Molecular Diversity (2005) 9: 131–139 c© Springer 2005

Full-length paper

Rational selection of structurally diverse natural product scaffolds with favorableADME properties for drug discovery

D.S. Samiulla1,†, V.V. Vaidyanathan1,†, P.C. Arun1, G. Balan1, M. Blaze1, S. Bondre1, G. Chandrasekhar1,A. Gadakh1, R. Kumar1, G. Kharvi1, H.-O. Kim2, S. Kumar1, J.A. Malikayil1, M. Moger1, M.K. Mone1,P. Nagarjuna1, C. Ogbu2, D. Pendhalkar1, A.V.S. Raja Rao1, G. Venkateshwar Rao1, V.K. Sarma1,S. Shaik1, G.V.R. Sharma1, S. Singh1, C. Sreedhar1, R. Sonawane1, U. Timmanna1 & L.W. Hardy2,∗1Aurigene Discovery Technologies, Ltd., Electronic City Phase 2, Hosur Road, Bangalore-562158, India; 2AurigeneDiscovery Technologies, Inc., 99 Hayden Avenue, Lexington, Massachusetts 02420, U.S.A.(∗Author for correspondence, E-mail: LHardy@charter.net, Tel: 508-347-5320, Fax: 978-685-3734)

Received 16 March 2004; Accepted 14 May 2004

Key words: computational design, early ADME, lead-like, metabolic stability, natural products, permeability, scaffold

Summary

Natural product analogs are significant sources for therapeutic agents. To capitalize efficiently on the effective features ofnaturally occurring substances, a natural product-based library production platform has been devised at Aurigene for druglead discovery. This approach combines the attractive biological and physicochemical properties of natural product scaffolds,provided by eons of natural selection, with the chemical diversity available from parallel synthetic methods. Virtual propertyanalysis, using computational methods described here, guides the selection of a set of natural product scaffolds that are bothstructurally diverse and likely to have favorable pharmacokinetic properties. The experimental characterization of several invitro ADME properties of twenty of these scaffolds, and of a small set of designed congeners based upon one scaffold, is alsodescribed. These data confirm that most of the scaffolds and the designed library members have properties favorable to theirutilization for creating libraries of lead-like molecules.

Abbreviations: ADME, absorption-distribution-metabolism-excretion; BCUT, Burden-Chemical Abstract Services – Universityof Texas; DMSO, dimethylsulfoxide; e-ADME, early ADME; HPLC/UV, high performance liquid chromatography moni-tored by ultraviolet absorbance; LC-MS, liquid chromatography–mass spectrometry; MDCK, Madin-Darby canine kidney;NCI, National Cancer Institute; PBS, phosphate buffered saline; PSA, polar surface area; TEER, transepithelial electricalresistance

Introduction

Natural products have historically provided important andeffective therapeutic agents [1]. Aspirin, penicillin, and pa-clitaxel are just three of numerous well-known natural prod-ucts used to improve human health. Moreover, analogs ofsuch natural products as benzylpenicillin and semisyntheticdrugs, such as taxotere, have proven to be enormously fruit-ful for new drug discovery. Recent studies have shown thatnatural compounds and their analogs continue to be majorsources of new drugs [2, 3]. Collections of natural productsexhibit physicochemical property profiles that are favorablecompared to those of drugs and complementary to those

†These authors have contributed equally to this work.

provided by synthetic compounds that derive from combi-natorial chemistry [4, 5].

Despite these advantages, the classical processes to iden-tify discrete new chemical entities from natural productsources are too inefficient to have survived in many of the cur-rent discovery environments at pharmaceutical companies.However, interest in natural products and their analogs assources of pharmaceutical agents has shown recent resilience[6–8]. To capitalize more efficiently on the effective featuresof naturally occurring substances, we have adopted a naturalproduct-based library production platform for lead discovery.This approach combines the attractive biological and physic-ochemical properties of natural product scaffolds, provided

132

by eons of natural selection, with the chemical diversity avail-able from parallel synthetic methods. The selection of naturalproduct scaffolds is guided by several criteria. Three majorcriteria are: (1) computational analyses of virtual propertiescalculated from the structures of the small molecule scaffoldsalone, (2) the density of patented entities in “chemical space”around the each scaffold being considered, and (3) accessi-bility (by either total synthesis or extraction from a renew-able source). The virtual property analysis considers both thediversity for the scaffolds set as a whole (e.g., how well thescaffolds complement each other in “chemical diversity”) andthe structural diversity accessible for each particular scaffold(either by total synthesis of analogs or by derivatization of ascaffold with multiple potential positions for modification).Further, the virtual property assessment includes a consid-eration of the “lead-like” nature of the scaffolds, so that agood starting point exists for lead generation and optimiza-tion. The goal of this process is to define a scaffold set thatis structurally diverse and contains members likely to havefavorable pharmacokinetic properties.

At present, fifty-one natural product scaffolds have beenselected, using the process broadly outlined above, for fur-ther exploration. Three examples are cryptoheptine, vasici-none, and Z-guggulsterone (Figure 1A). Z-Guggulsterone,the active constituent of an ayurvedic medicine used in In-dia for several thousand years, is especially exciting. Z-Guggulsterone has clinically proven benefits in the treatmentof human hyperlipidemia, and is approved and marketed forthat purpose in India [9]. It was recently proven, using geneknockout methods, that Z-guggulsterone targets the farnesoidX receptor (also known as the bile acid receptor) in its effectson cholesterol metabolism in the mouse [10].

(A)

(B)

Figure 1. (A) Representative natural product scaffolds, (B) Representative analogs of vasicinone.

The production of small molecule libraries based onthese natural product scaffolds is cost-effective, efficient,expedient, and amenable to solid-phase and parallel solution-phase synthesis. For example, a small library of vasicinoneanalogs was chosen for synthesis from a virtual library to pro-vide a structurally diverse set with acceptable ADME param-eters. Several members of the vasicinone library are shownin Figure 1B1. The computational analysis of the vasicinonelibrary is described here. For target-driven drug discovery,the designs of small molecule libraries based on the naturalproduct scaffolds are ideally guided not by diversity met-rics primarily, but by the atomic structures of the targets.Design of our targeted libraries is based both on in silicoestimates of the ADME properties of the virtual librariescalculated solely from the small molecule structures and bycomputational screening of virtual libraries against structuralmodels of specific proteins that are targeted for therapeuticdiscovery.

Major reasons for the failure of early drug candidates toreach the market include inappropriate ADME (absorption,distribution, metabolism and excretion) properties of the can-didates, and metabolism-based adverse interactions of thecandidates with existing drugs. By some estimates, 40–50%of new chemical entities in the drug discovery pipeline faildue to poor ADME properties [11, 12]. From a commercialperspective, poorly behaved compounds need to be recog-nized and removed as early as possible in the discovery pro-cess, rather than during the more costly clinical testing anddevelopment phase. To speed the provision of the data neededfor the design and discovery of well-behaved drug candi-dates, “early ADME” (e-ADME) assays have been developed[13–15]. These assays use in vitro surrogates for in vivo

133

physiology, and allow information about ADME andmetabolism-based adverse drug-drug interactions to be in-corporated early in drug discovery. The e-ADME parame-ters of twenty of our natural product scaffolds have beencharacterized. The properties studied were aqueous solubil-ity at pH 7.4, inhibition of several recombinant human cy-tochrome P450 isozymes, stability to exposure to pooled hu-man hepatic microsomes, and permeability across the MDCK(Madin-Darby Canine Kidney) cell monolayer. Incorporationof early ADME assessments in our drug discovery projectshas helped us to identify natural product scaffolds and nat-ural product analogs with high probability of good intesti-nal absorption and to flag potential problems arising fromhepatic metabolism. This approach allows potential prob-lems with poor absorption or metabolic interactions to beavoided by re-design of analogs at the discovery stage in acost-effective way.

Methods

Computational property and diversity analyses

The BCUT parameter approach, pioneered by Pearlman [16],was employed as provided in the software package fromTripos [17]. To establish statistically valid orthogonal axesof the chemistry vector space to use for the diversity analysisof natural product scaffolds, BCUT parameters were gener-ated for about 25,000 molecules from the National CancerInstitute (NCI) database, as provided by Tripos. Correlationbetween the axes was calculated and a chemistry space wasdefined with minimal r2 (<0.25) values between the axes.The diversity of scaffold molecular structures was comparedin this chemistry space based on partitioning (cell based)methods.

Evaluation of the vasicinone library for diversity was donedifferently from the evaluation of the scaffolds. Relative di-versity is important for a library with respect to the scaffold[18]. Therefore, the BCUTs for the members of the vasi-cinone library were generated. Chemistry space definitionswere generated which maintained the r2 (between the axes)less than 0.25. As the number of library members was low,it would not be statistically valid to assume the axis to beorthogonal. Hence, the same BCUTs were generated for theNCI set of molecules, and r2 between the axes were eval-uated. The chemistry space with r2 (between the axes) lessthan 0.25 was then accepted, and was partitioned into cellssuch that the occupancy was 14.8%. Statistics of occupancyfor the compounds were calculated. The chemistry spaces inboth the cases are described as: two partial charges basedBCUTs, one H-donor based BCUT, one H- acceptor basedBCUT, and two polarizability based BCUTs.

A molecular fingerprint-based analysis was also per-formed for both the scaffolds and the vasicinone library mem-bers, using 992 bit Unity 2D fingerprint definitions. Thefingerprints were generated, and the correlations between

the various structural fragments present were calculated [19].The Tanimoto similarity index was defined as:

T ( f1, f2) = N1&2/N1|2

where, f1 = fingerprint 1, f2 = fingerprint 2, N1&2 = num-ber of fields present in both f1 and f2, N1|2 = number offields present in either f1 or f2.

A variety of structure-based parameters, relevant toADME characteristics such as those described by Lipinski[20] and by Veber and co-workers [21], were also evaluatedfor the compounds. The parameters calculated were ClogP(calculated using Sybyl from Tripos), molecular weight andnumber of rotatable bonds (assessed manually), and num-ber of hydrogen bond donors and acceptors (assessed usingChemDraw for Excel). In the calculation of the number ofrotatable bonds, amide bonds are not counted, since the ro-tational barrier should prevent free rotation. The polar sur-face area (PSA) was calculated using the MolCad module ofSybyl.

General analytical methods

Fluorescence measurements were done in 96-well platesusing a Spectromax Gemini XS reader (Molecular De-vices, USA). Transepithelial resistance (TEER) measure-ments were made using a World Precision Instrument(Florida, USA) probe. HPLC analyses (performed on anAgilent Technologies 1100 series system) were done usinga reverse phase column (Zorbax Eclipse XDB C18, 150 ×4.6 mm, 3.5 micron), with quantitation based on UV ab-sorbance at 254 nm. The LC-MS analyses were performed inmultiple reaction monitoring mode using an Applied Biosys-tems API 2000, fed by a in-line Agilent Technologies 1100series HPLC using the same model of Zorbax reverse phasecolumn for sample chromatography.

Aqueous solubility assay

DMSO solutions of test compounds were added to phosphatebuffered saline pH 7.4 (PBS) in a 96 deep-well plate to gen-erate theoretical concentrations of 200 µM in 2% DMSO.The solutions were equilibrated by shaking (200 rpm, Ikaplate shaker) for 16 h at 25 ◦C. Undissolved compound wasremoved by centrifugation, and the supernatant was analyzedby HPLC-UV or LC/MS/MS. Aqueous solubility was calcu-lated using the equation:

Aqueous solubility = 200 µM × PAPBS/PADMSO

where PAPBS and PADMSO are the peak areas from the anal-yses of test compound in PBS with 2% DMSO and of testcompound in 100% DMSO, respectively.

Assays were performed in duplicate or triplicate.

134

MDCK monolayer permeability assay

Madin-Darby Canine Kidney (MDCK) cells (American TypeCulture Collection, Manassas, Virginia, USA) were used toassess compound permeability [22]. The cells were grown inDulbecco’s modified Eagle medium supplemented with 10%fetal bovine serum, 1 mM non-essential amino acids, 1 mMsodium pyruvate and gentamicin sulphate (50 µg/mL) to 70–80% confluency prior to seeding in 24-well plates loaded withpolycarbonate Millicell inserts (Millipore, 12 mm diameter,0.4 µm, 50,000 cells/insert). Cells were fed with freshmedium every other day and incubated at 37 ◦C, with 5% CO2for 3 days prior to drug transport experiment. Cell monolayerintegrity was assessed by measuring TEER values. Drugswere applied at 50 µM in Hank’s buffered salt solution con-taining 0.5% DMSO to the apical chamber and the transportassay was carried out for 2 h at 37 ◦C. At the end of theassay, samples from both apical and basal chambers werecollected for analysis by HPLC/UV or LC/MS/MS, and themonolayer integrity was re-assessed by dye rejection usingLucifer Yellow. Data from wells with dye rejection less than98% were rejected. Apparent permeability (Papp) values werecalculated using the equation:

Papp = d Q/dt × 1/Co × 1/A

where, dQ/dt = permeability rate in µg/sec, Co = initial con-centration in µg/mL, A = membrane surface area (0.6 cm2

for 12 mm inserts), Papp values were expressed in cm/sec.All the assays were performed in triplicate.

Human hepatic microsomal stability assay

Pooled human liver microsomes were employed to assess thepotential instability of compounds to phase I metabolism [23,24]. The microsomes (BD Gentest) were incubated with testcompounds (1–5 µM in 0.2% DMSO 1 buffer) at 37 ◦C for30 min. The reaction was stopped by the addition of acetoni-trile containing haloperidol as internal standard. Precipitatedprotein was removed by centrifugation and the supernatantswere analyzed by HPLC-UV or LC/MS/MS method. Stabil-ity was assessed by the disappearance of compound based onthe change in analyte to internal standard peak height ratio.Metabolic stability was defined as the amount of test com-pound remaining after the incubation with human hepaticmicrosomes, and expressed as a percentage of the initial con-centration. All assays were performed in triplicate.

Human cytochrome P450 inhibition assay

Human recombinant Cytochrome P450 isozymes 2C9, 2D6and 3A4 (BD Gentest) were incubated with 5 µM of testcompound in buffer containing 0.1% DMSO for 10 min at37 ◦C. The residual enzyme activity was measured [15] usingthe following fluorogenic substrates (also obtained from

BD Gentest): 7-methoxy-4-(trifluoromethyl) coumarin forCYP2C9, 3[2-N,N-diethyl-N-methylammonium)ethyl]-7-methoxy-4-methylcoumarin for CYP2D6, and 7-benzyloxy-4-(trifluoromethyl)coumarin for CYP3A4. All assays wereperformed in duplicate.

Results and discussion

To understand chemical diversity for sets of compoundsto be used for chemical biology, that diversity must bequantified in a manner relevant to the biological role forthe compounds. The methods used to quantify diversity are“fingerprint” based approaches and “chemistry space” basedapproaches. In fingerprint-based approaches, bit strings

Figure 2. Distribution of the cell occupancy for the vasicinone librarymembers.

Figure 3. Scatter plot of molecular weight, ClogP, and PSA for naturalproduct scaffolds.

135

(molecular fingerprints) that indicate the presence of sub-structural features in each molecule are employed. Similarityand dissimilarity between two molecules are measured asthe inner product of these bit string vectors, or a closelyrelated quantity called Tanimoto similarity/dissimilarityindices [19]. This kind of approach is helpful in designingcombinatorial libraries as it checks for presence of differentkinds of fragments. However, with a collection of naturalproduct scaffolds, this method is not relevant for defining alibrary, which is diverse in a general sense.

The chemistry space approach relies on the definitionof a set of quantifiable chemical descriptors, corresponding

Figure 4. Scatter plot of number of rotatable bonds, H bond acceptors, andH bond donors for natural product scaffolds.

Figure 5. Scatter plot of molecular weight, ClogP, and PSA for the vasici-none library.

to dimensions within chemistry space. As the position ofa particle in three dimensions is defined by its x-, y-, andz-co-ordinates, the positions of molecules in N-dimensionalchemical space are defined by N chemical descriptors. A sim-ilar approach was exemplified by Pearlman and co-workersin their analysis of acetylcholinesterase inhibitors [16]. Theaxes of the chemistry space are BCUTs, which are eigen-values of the connectivity matrices of the molecular graphswith some properties as the diagonal elements. Pearlman andco-workers found that three of the six BCUT metrics, theyhad identified from an analysis of a database of 60,000 ‘drug-like’ compounds, were receptor relevant, i.e. the actives wereclustered in these dimensions. We have employed this sameapproach to identify scaffolds that are diverse in a biologi-cally relevant chemical space, as defined by the diversity inthe compounds in the NCI database.

The natural product scaffolds are diverse in the chemicalspace, since the different scaffolds are found in different cells.

Figure 6. Scatter plot of number of rotatable bonds, H bond donors, and Hbond acceptors for the vasicinone library.

Figure 7. Aqueous solubility of standard drugs in solubility assay correlatedwith published data.

136

As the distribution of partial charges, polarizability, and H-bonding properties of the structures are mapped by BCUTsbased upon them, these scaffolds are diverse in the distribu-tion of these properties. The variety of distribution of theseproperties in the natural product scaffolds gives them a betterchance to be selected by varied types of receptors.

The bits in the fingerprints indicate different structuralfragments present in the scaffolds. The fact that only 62bits were highly correlated out of 992 indicates that thesescaffolds have different chemical cores in them. Thus, thefingerprints based approach indicates a statistically signifi-cant difference in the types of chemical classes present in thescaffolds. This conforms to the chemically intuitive definition

Figure 8. Permeability in MDCK model system correlates well with known human absorption data for orally dosed drugs.

Figure 9. Cytochrome P450 inhibition assay with standard inhibitors.

of the diversity of natural product scaffolds [4, 5]. Not sur-prisingly, the vasicinone library was considerably less diversethan the Aurigene collection of natural product scaffolds, withmultiple compounds in many cells (cf., Figure 2).

As shown in Figures 3 and 4, the Lipinski parametersfor the natural product scaffolds satisfy “the rule of five”limits [20]. The number of rotatable bonds is less than sevenfor all the molecules, which is well within the maximum of10 suggested for good bioavailability [21]. The maximumPSA for these scaffolds is 310 A2. The acceptable maximumfragment-based PSA is 140 A2, compounds with PSAvalues above this limit are expected to have low oralbioavailability [14]. In our experience, the PSA calculated

137

Table 1. ADME properties of natural product scaffolds

Permeability throughSolubility at pH MDCK monolayer (Papp),

Scaffold 7.4, 10−6 M 10−6 cm/s (±SD)

1 (Vasicinone) >200 74 (±3)

2 (Cryptoheptine) insoluble nd

3 40 6.5 (±0.8)

4 89.7 60 (±2.4)

5 196.2 7.5 (±0.2)

6 (Z-Guggulsterone) 94 33.1 (±0.7)

7 195.9 68 (±4)

8 >200 83 (±3)

9 1.9 nd

10 191 19.1 (±0.4)

11 80 79 (±7)

12 >200 62 (±4)

13 197.5 113 (±8)

14 8.7 70.9 (±0.1)

15 >200 69 (±8)

16 6.1 nd

17 178.5 57 (±3)

18 >200 5.9 (±0.5)

19 >200 46 (±4)

20 13.5 10.5 (±0.8)

Values are mean (±SD) of triplicate assays. Green = Good, Blue = Ac-ceptable, Red = May need to be modified by analog synthesis. nd = Notdetermined.

using the MolCad method with the energy minimizedstructures (as implemented by Tripos) is 2.5 times the valuecalculated by the faster fragment-based approach [26].Therefore, the acceptable limit for the PSA value computedwith MolCad is 350 A2. Thus, there are no computationalalerts for these molecules for solubility, permeability, andoral bioavailability.

Table 2. ADME properties of Vasicinone analogs

Vasicinone analogs Solubility, 10−6 M Papp, 10−6 cm/s 3A4 % inhibition 2C9 % inhibition 2D6 % inhibition

1 >200 69 (±8) 34 43 14

2 >200 87 (±4) 34 26 33

3 >200 67 (±2) 32 25 16

4 >200 75 (±4) 30 27 19

5 >200 46 (±2) 31 22 24

6 >200 65 (±2) 31 32 6

7 >200 64 (±17) 31 29 19

8 >200 73 (±5) 31 36 23

9 >200 35 (±2) 49 66 16

10 >200 2.2 (±0.6) 31 44 7

11 11.6 nd 32 32 17

12 >200 97 (±1) 26 36 24

Green = Good, Blue = Acceptable, Red = requires modification by analog synthesis. nd = Not determined.

Distributions of the ADME relevant computed proper-ties for the vasicinone library are shown in Figures 5 and 6.Among the 172 molecules in the vasicinone library, there arefive molecules with molecular weight greater than 500, andone molecule has a ClogP value greater than 5. The Mol-Cad computed PSA has a maximum value of 311 A2 for thelibrary. The maximum values observed for the numbers ofH-Bond donors and H-Bond acceptors are 4 and 6, respec-tively. The maximum number of rotatable bonds for the anymember of the library is 10. Therefore, six molecules havecomputational alerts.

The experimental assays for the e-ADME parameterswere validated using drugs and other well characterized com-pounds with known values for solubility (Figure 7), per-meability (Figure 8), inhibitory potency toward cytochromeP450s (Figure 9), and metabolic stability (Figure 10). Thesevalidated in vitro assays were then used to assess the e-ADME parameters (solubility, permeability, inhibition of cy-tochrome P450 isozymes, and hepatic microsomal stability)for twenty of the natural product scaffolds, and for a subset ofcompounds in the vasicinone library. Reference compoundswith low, medium, and high values for the assayed prop-erty are routinely included during the assessments of testcompounds to assure the acceptable performance of theseassays.

Most of the scaffolds and vasicinone library members hadacceptable or good aqueous solubility and permeability pa-rameters (Tables 1 and 2), which supports the accuracy ofthe favorable computational estimates of their physicochem-ical properties. The majority of the natural product scaffoldshave excellent microsomal stability, with more than 50% ofthe parent compound surviving the 30-min exposure for 18of the 20 scaffolds examined (Figure 11). Most of the naturalproduct scaffolds and vasicinone analogs also have accept-ably weak or moderate inhibitory potencies towards the threecytochrome P450 isozymes that were assayed (Table 2 andFigure 12), which are the prevalent hepatic P450 isozymes in

138

Figure 10. Microsomal stability assay with standard compounds.

Figure 11. Microsomal stability of natural products scaffolds.

Figure 12. Inhibition of CYP450 isozymes by natural product scaffolds.

139

most humans [27]. Generally, the natural product scaffoldshave good e-ADME properties. This strongly supports theirsuitability for library generation for drug discovery. Preser-vation of the favorable e-ADME properties during librarybuildouts starting with the scaffolds is suggested by the goode-ADME properties exhibited by the vasicinone library mem-bers. A few scaffolds have marginal e-ADME properties, in-dicating that if they are pursued, these properties will likelyrequire modification during library build-outs.

In summary, the computational and experimental assess-ments of the properties of the natural product scaffolds arein agreement. These findings support the utility of the ratio-nal process that we have established to choose scaffolds forhit and lead generation libraries for drug discovery. Compu-tational methods clearly have great utility to combine goodphysicochemical properties (and thereby increase the likeli-hood of favorable bioavailability) and structural diversity inlibraries of natural product analogs. We are pursuing a sim-ilar approach in the design, synthesis, and characterizationof small, target-focused libraries of natural product analogswith good physicochemical properties, guided by computa-tional methods and the structural templates of therapeutictarget proteins.

Note

1. The chemical structures of most of the scaffolds and vasicinone librarymembers cannot be disclosed for proprietary reasons.

References

1. Samuelsson, G., Drugs of natural origin, Swedish Pharmaceutical So-ciety, Stockholm, Sweden, 1991.

2. Buss, A.D., Cox, B. and Waigh, R.D., Natural products as leads for newpharmaceuticals, In Abraham, D.J. (Ed), Burger’s Medicinal Chem-istry and Drug Discovery, Sixth Edition, Vol. 1, Wiley Interscience,New York, 2003, pp. 847–900.

3. Newman, D.J., Cragg, G.M. and Snader, K.M., Natural products assources of new drugs over the period 1981–2002, J. Nat. Prod., 66(2003) 1022–1037.

4. Feher, M. and Schmidt, J.M., Property distributions: Differences be-tween drugs, natural products, and molecules from combinatorialchemistry, J. Chem. Inf. Comput. Sci., 43 (2003) 218–227.

5. Henkel, T., Brunne, R.M., Muller, H. and Reichel, F., Statistical inves-tigation into the structural complementarity of natural products andsynthetic compounds, Angew. Chem. Int., 38 (1999) 643–647.

6. Rouhi, A.M., Rediscovering natural products, Chem. Eng. News, 81(2003) 77–91.

7. Rouhi, A.M., Betting on natural products for cures, Chem. Eng. News,81 (2003) 93–103.

8. Rouhi, A.M., Moving beyond natural products, Chem. Eng. News, 81(2003) 104–107.

9. Urizar, N.L. and Moore, D.D., Gugulipid: A natural cholesterol-lowering agent, Annu. Rev. Nutr., 23 (2003) 303–313.

10. Urizar, N.L., Liverman, A.B., Dodds, D.T., Silva, F.V., Ordentlich, P.,Yan, Y., Gonzalez, F.J., Heyman, R.A., Mangelsdorf, D.J. and Moore,D.D., A natural product that lowers cholesterol as an antagonist ligandfor FXR, Science, 296 (2002) 1703–1706.

11. van de Waterbeemd, H. and Gifford, E. ADMET in silico modelling: To-wards prediction paradise? Nat. Rev. Drug. Discov., 2 (2003) 192–204.

12. Kennedy, T., Managing the drug discovery/development interface,Drug Discov. Today, 2 (1997) 436–444.

13. Stoner, C.L., Cleton, A., Johnson, K., Oh, D.M., Hallak, H.,Brodfuehrer, J., Surendran, N. and Han, H.K., Integrated oralbioavailability projection using in vitro screening data as a se-lection tool in drug discovery, Int. J. Pharm., 268 (2004) 241–249.

14. Yu, H., and Adedoyin, A., ADME-Tox in drug discovery: Integrationof experimental and computational technologies, Drug Discov. Today,8 (2003) 852–861.

15. Crespi, C.L., Miller, V.P. and Stresser, D.M., Design and application offluorometric assays for human cytochrome P450 inhibition, MethodsEnzymol., 357 (2002) 276–284.

16. Pearlman, R.S. and Smith, K.M., Metric validation and the receptor-relevant subspace concept, J. Chem. Inf. Comput. Sci., 39 (1999)28–35.

17. Diverse Solutions Users Manual version 3.0.1: Laboratory ofMolecular Graphics and Theoretical Modeling, University of Texas,Austin,Texas copyrighted 1996, 1997, Distributed by Tripos Inc.,1699 S. Hanley Road, St. Louis, MO 63144 (1996, 1997).

18. Mason, J.S., Absolute versus relative similarity and diversity, InDean, P.M., and Lewis, R.A. (Eds ) Molecular Diversity in DrugDesign, Kluwer Academic Publishers, Dordrecht (1999), pp.88–90.

19. Holliday, J.D., Ranade, S.S. and Willett, P., A fast algorithm forselecting sets of dissimilar molecules from large chemical databases,Quant. Struct.-Act. Relat., 14 (1995) 501–506.

20. Lipinski, C.A., Drug-like properties and the causes of poor solubilityand poor permeability, J. Pharmacol. Toxicol. Methods, 44 (2000)235–249.

21. Veber, D.F., Johnson, S.R., Cheng, H.Y., Smith, B.R., Ward, K.W. andKopple, D.K., Molecular properties that influence oral bioavailabilityof drug candidates, J. Med. Chem., 45 (2002) 2615–2623.

22. Irvine, J.D., Takahashi, L., Lockhart, K., Cheong, J., Tolan, J.W.,Selick, H.E. and Grove, J.R., MDCK (Madin-Darby canine kidney)cells: A tool for membrane permeability screening, J. Pharm. Sci., 88(1999) 28–33.

23. Iwatsubo, T., Hirota, N., Ooie, T., Suzuki, H., Shimada, N., Chiba, K.,Ishizaki, T., Green, C. E., Tyson, C.A. and Sugiyama, Y., Prediction ofin vivo drug metabolism in the human liver from in vitro metabolismdata. Pharmacol. Ther., 73 (1997) 147–171.

24. Mandagere, A.K., Thompson, T.N. and Hwang, K.K., Graphicalmodel for estimating oral bioavailability of drugs in humans andother species from their Caco-2 permeability and in vitro liverenzyme metabolic stability rates, J. Med. Chem., 45 (2002) 304–311.

25. Palm, K., Stenberg, P., Luthman, K. and Artursson, P., Polar molec-ular surface properties predict the intestinal absorption of drugs inhumans, Pharm. Res., 14 (1997) 568–571.

26. Clark, D.E., Rapid calculation of polar molecular surface areaand its application to the prediction of transport phenomena. 1.Prediction of intestinal absorption, J. Pharm. Sci., 88 (1999) 807–814.

27. Yan, Z. and Caldwell, G.W., Metabolism profiling, and cytochromeP450 inhibition and induction in drug discovery, Curr. Top. Med.Chem., 1 (2001) 403–425.