Save (0)




Cheng Chang and Peter W. Swaan


20.1 Introduction
20.2 Modeling Techniques
20.3 Drug Absorption
20.3.1 Solubility
20.3.2 Intestinal Permeation
20.3.3 Other Considerations
20.4 Drug Distribution
20.5 Drug Excretion
20.6 Active Transport
20.6.1 P-gp
20.6.2 BCRP
20.6.3 Nucleoside Transporters
20.6.4 hPEPT1
20.6.5 ASBT
20.6.6 OCT
20.6.7 OATP
20.6.8 BBB-Choline Transporter
20.7 Current Challenges and Future Directions

Computer Applications in Pharmaceutical Research and Development, Edited by Sean Ekins.
ISBN 0-471-73779-8 Copyright © 2006 John Wiley & Sons, Inc.





Historically, drug discovery has focused almost exclusively on effi cacy and
selectivity against the biological target. As a result, nearly half of drug can-
didates fail at phase II and phase III clinical trials because of undesirable
drug pharmacokinetics properties, including absorption, distribution, metab-
olism, excretion, and toxicity (ADMET). The pressure to control the escalat-
ing cost of new drug development has changed the paradigm since the
mid-1990s. To reduce the attrition rate at more expensive later stages, in vitro
evaluation of ADMET properties in the early phase of drug discovery has
been widely adopted. Many high-throughput in vitro ADMET property
screening assays have been developed and applied successfully [1]. For
example, Caco-2 and MDCK cell monolayers are widely used to simulate
membrane permeability as an in vitro estimation of in vivo absorption. These
in vitro results have enabled the training of in silico models, which could be
applied to predict the ADMET properties of compounds even before they
are synthesized. Fueled by the ever-increasing computational power and sig-
nifi cant advances of in silico modeling algorithms, numerous computational
programs that aim at modeling drug ADMET properties have emerged. A
comprehensive list of available commercial ADMET modeling software has
been provided previously by van de Waterbeemd and Gifford [2].

Our discussion in this chapter focuses on in silico modeling of drug dis-
position including absorption, distribution, and excretion (Fig. 20.1). We
begin with a summary of in silico techniques in modeling drug ADMET
properties, followed by a discussion of current progress in modeling different
aspects of drug disposition at the systemic level. Recent advancements in
modeling a diverse array of active transporters as well as their impact on
drug pharmacokinetic profi les are also reviewed. This chapter concludes
with the challenges and future trends of in silico drug disposition property


There are mainly two types of modeling approaches. The quantitative
approaches represented by pharmacophore modeling and fl exible docking
studies investigate the structural requirements for the interaction between
drugs and the targets that are involved in ADMET processes. These are
especially useful when there is an accumulation of knowledge against a certain
target. For example, a set of drugs known to be transported by a transporter
would enable a pharmacophore study to elucidate the minimum required
structural features for transport. The availability of a protein’s three-
dimensional structure, from either X-ray crystallization or homology model-
ing, would assist fl exible docking of the active ligand to derive important
interactions between the protein and the ligand. Three widely used automated



Figure 20.1 In silico modeling targets of drug disposition.



pharmacophore perception tools, DISCO (DIStance COmparisons) [3],
GASP (Genetic Algorithm Similarity Program) [4], and Catalyst/HIPHOP
[5], were critically evaluated and compared by Patel and colleagues [6]. All
three programs attempt to determine common features based on the super-
position of active compounds with different algorithms. The application of
different fl exible docking algorithms in drug discovery has recently been
reviewed [7]. The essential interactions derived from either study can be used
as a screen in evaluating drug ADMET properties.

The qualitative approaches represented by quantitative structure-activity
relationship (QSAR) and quantitative structure-property relationship (QSPR)
studies utilize multivariate analysis to correlate molecular descriptors with
ADMET-related properties. A diverse range of molecular descriptors can be
calculated based on the drug structure. Some of these descriptors are closely
related to a physical property and are easy to comprehend (e.g., molecular
weight), whereas the majority of the descriptors are of quantum mechanical
concepts or interaction energies at dispersed space points that are beyond
simple physicochemical parameters. When calculating correlations, it is
important to select the molecular descriptors that represent the type of inter-
actions contributing to the targeted biological property. In fact, a set of
descriptors that specifi cally target ADME related properties has been pro-
posed by Cruciani and colleagues [8]. The majority of published ADMET
models are generated based on 2D descriptors. Even though the alignment-
dependent 3D descriptors that are relevant to the targeted biological activity
tend to generate the most predictive models, the diffi culties inherent in
structure alignment thwart attempts to apply this type of modeling in a
high-throughput manner. This has prompted the development of alignment-
independent 3D descriptors. However, most of these descriptors to date are
still insuffi ciently discriminating.

A wide selection of statistical algorithms is available to researchers for
correlating fi eld descriptors with ADMET properties including simple mul-
tiple linear regression (MLR), multivariate partial least-squares (PLS), and
the nonlinear regression-type algorithms such as artifi cial neural networks
(ANN) and support vector machine (SVM). No one method can consistently
perform better than the others. Just like descriptor selection, it is essential to
select the right mathematical tool for most effective ADMET modeling.
Sometimes it is necessary to apply multiple statistical methods and compare
the results to identify the best approach, as illustrated in a recent solubility
QSPR model [9].


Because of its convenience and good patient compliance, oral administration
is the most preferred drug delivery form. As a result, much of the attention
of in silico approaches is focused on modeling drug oral absorption, which
mainly occurs in the human intestine. In general, drug bioavailability and



absorption is the result of the interplay between drug solubility and intestinal

20.3.1 Solubility

A drug generally must dissolve before it can be absorbed from the intestinal
lumen. Direct measurement of solubility is time-consuming and requires a
large amount of (expensive) compound at the milligram scale. By measuring
a drug’s logP value (log of the partition coeffi cient of the compound between
water and n-octanol) and its melting point, one could indirectly estimate solu-
bility using the “general solubility equation” [10]. Even though the process is
simplifi ed, it still requires the synthesis of the compound. To predict the solu-
bility of the compound even before synthesizing it, in silico modeling can be
implemented. There are mainly two approaches to modeling solubility. One
is based on the underlying physiological processes, and the other is an empiri-
cal approach.

The dissolution process involves the breaking up of the solute from its
crystal lattice and the association of the solute with solvent molecules. Obvi-
ously, weaker interactions within the crystal lattice (lower melting point) and
stronger interactions between solute and solvent molecules will result in better
solubility and vice versa. For druglike molecules, solvent-solute interaction
has been the major determinant of solubility and its prediction attracts most
efforts. LogP is the simplest estimation of solvent-solute interaction and can
be readily predicted with commercial programs such as CLogP (Daylight
Chemical Information Systems, Aliso Viejo, CA), which utilizes a fragment-
based approach. To recognize the contribution of solute crystal lattice energy
in determining solubility, other approaches amended LogP values with addi-
tional terms for more accurate predictions [11, 12].

Empirical approaches, represented by QSPR, utilize multivariate analyses
to identify correlations between molecular descriptors and solubility. Even
though the calculation process ignores the underlying physiological processes,
the molecular descriptor selection and model interpretation still requires
understanding of the dissolution process. Selection of fi eld descriptors that
adequately describe the physiological process and the appropriate multivari-
ate analysis is essential to successful modeling. The target property for most
models is the logarithm of solubility (logS), and many models are trained and
verifi ed with the AQUASOL (
aquasol/) and PhysProp ( data-
bases. Lombardo and colleagues have provided a critical review of available
solubility prediction algorithms [13].

20.3.2 Intestinal Permeation

Intestinal permeation describes the ability of drugs to cross the intestinal
mucosa separating the gut lumen from the portal circulation. It is an essential



process for drugs to pass the intestinal membrane before entering the sys-
temic circulation to reach their target site of action. The process involves both
passive diffusion and active transport. It is a complex process that is diffi cult
to predict solely based on molecular mechanism. As a result, most current
models aim to simulate in vitro membrane permeation of Caco-2, MDCK
[14], or PAMPA [15], which have been a useful indicator of in vivo drug
absorption [16, 17]. The current progress of intestinal permeation research
has been reviewed by Malkia and colleagues [18].

20.3.3 Other Considerations

The ionization state will affect both solubility and permeability and, as a
result, infl uence the absorption profi le of a compound. Given the environ-
mental pH, the charge of a molecule can be determined using the compound’s
ionization constant value (pKa), which indicates the strength of an acid or
a base. Several commercially and publicly available programs provide pKa
estimation based on the input structure, including SCSpKa (ChemSilico,
Tewksbury, MA), Pallas/pKalc (CompuDrug, Sedona, AZ), ACD/pKa
(ACD, Toronto, ON, Canada), and SPARC online calculator (http://ibmlc2.

Both infl ux and effl ux transporters are located in intestinal epithelial cells
and can either increase or decrease oral absorption. Infl ux transporters such
as human peptide transporter 1 (hPEPT1), apical sodium bile acid trans-
porter (ASBT), and nucleoside transporters actively transport drugs that
mimic their native substrates across the epithelial cell, whereas effl ux trans-
porters such as P-glycoprotein (P-gp), multidrug resistance-associated protein
(MRP), and breast cancer resistance protein (BCRP) actively pump absorbed
drugs back into the intestinal lumen.

To correctly predict overall oral absorption, drug metabolism in intestinal
epithelial cells by cytochrome P450 enzymes should also be considered. The
prediction of drug metabolism has already been covered in detail in Chapter

Other than the different approaches mentioned above, commercial pack-
ages such as GastroPlus (Simulations Plus, Lancaster, CA) [19] and iDEA
(LionBioscience, Inc. Cambridge, MA) [19] are available to predict oral
absorption and other pharmacokinetic properties. They are both based on the
advanced compartmental absorption and transit (CAT) model [20], which
incorporates the effects of drug moving through the gastrointestinal tract and
its absorption into each compartment at the same time (see also Chapter 22).


Distribution is an important aspect of a drug’s pharmacokinetic profi le. The
structural and physiochemical properties of a drug determine the extent of



its distribution, which is mainly refl ected by three parameters: volume of
distribution (VD), plasma-protein binding (PPB), and blood-brain barrier
(BBB) permeability. VD is a measure of relative partitioning of drug between
plasma and tissue, an important proportional constant that, when combined
with drug clearance, could be used to predict drug half-life. The half-life of
a drug is a major determinant of how often the drug should be administered.
However, because of the scarcity of in vivo data and the complexity of the
underlying processes, computational models that are capable of predicting VD
based solely on computed descriptors are still under development. However
Lombardo and colleagues have proposed an approach to predicting VD for
neutral and basic compounds with two in vitro physicochemical parameters
[21]. With additional data, this model was further expanded and the robust-
ness of the approach was tested and validated [22]. This represents a step in
the right direction in accurately predicting VD.

Drugs bind to a variety of plasma proteins such as serum albumin. As
unbound drug primarily contributes to pharmacological effi cacy, the effect of
PPB is an important consideration when evaluating the effective (unbound)
drug plasma concentration. Several models have been proposed to predict
PPB [23–27]. As suggested by Lombardo and colleagues [13], the model
should not rely on the binding data of only one protein when predicting
plasma protein binding because it is a composite parameter refl ecting interac-
tions with multiple proteins. Recently, Yamazaki and Kanaoka applied a
nonlinear regression analysis over 300 drugs with experimental human PPB
percent data. For neutral and basic drugs they found a sigmoidal correlation
between logD (distribution coeffi cient) and PPB, and for acidic drugs the
same sigmoidal correlation between logP and PPB. The model was validated
with an external test set of 20 compounds. This work provides a useful
approximation of PPB.

The BBB maintains the restricted extracellular environment in the central
nerve system (CNS). The evaluation of drug penetration through the BBB is
an integral part of the drug discovery and development process. For drugs
that target the CNS, it is imperative they cross the BBB to reach their targets.
Conversely, for drugs with peripheral targets, it is desirable to restrict their
passage through the BBB to avoid CNS side effects. Again, because of the
few experimental data derived from inconsistent protocols, most BBB perme-
ation prediction models are of limited practical use despite intensive efforts
[28–32]. Most approaches model log blood/brain (logBB), which is a mea-
surement of the drug partitioning between blood and brain tissue. This mea-
surement is an indirect implication of the BBB permeability, which does not
discriminate between free and plasma protein-bound solute [33]. Pardridge
suggests modeling of a more accurate parameter, log BBB permeability-
surface area (logPS), which refl ects the free drug level in brain [33]. This new
concept was successfully adopted in two recent modeling studies [34, 35]. A
recent review discusses key considerations for development and application
of the BBB modeling [36]. In addition to forming complex tight junctions, the



presence of effl ux transporters and metabolic enzymes is another mechanism
that the BBB employs to prevent xenobiotics from entering the CNS. Three
types of drug effl ux transporters have been identifi ed from brain: multidrug
resistance transporters, monocarboxylic acid transporters, and organic ion
transporters. A large number of commonly prescribed drugs fall into the
categories of substrates of these effl ux transporters [37]. Failing to consider
these active transport systems would greatly compromise accuracy of the
BBB penetration prediction. Extensive substrate requirement studies have
been performed for multidrug resistance transporters, especially P-gp,
because of their infl uence on various aspects of drug discovery and develop-
ment. The role of monocarboxylic acid transporters and organic ion trans-
porters in the BBB is just being established through accumulating experimental
evidence, and no computational models have been generated to date. We can
expect to see such models with the accumulation of experimental data.


The excretion or clearance of a drug is quantifi ed by plasma clearance, which
is defi ned as plasma volume that has been cleared completely free of drug per
unit of time [38]. Together with VD, it can assist in the calculation of drug
half-life, thus determining dosage regime. Hepatic and renal clearances are
the two main components of plasma clearance. No model has been reported
that is capable of predicting plasma clearance solely from computed drug
structures. Current modeling efforts are mainly focused on estimating in vivo
clearance from in vitro data [39, 40]. Just like other pharmacokinetic aspects,
the hepatic and renal clearance process is also complicated by the presence
of active transporters. In a study performed by Sasaki and colleagues [40],
the effect of active transport is incorporated by measuring in vitro data from
MDCK cells that express organic anion transporting polypeptide (OATP) 4
and MRP2. However, to predict clearance for a given structure, knowledge
of the structural requirements for these transporters is required.


Transporters should be an integral part of any ADMET modeling program
because of their ubiquitous presence on barrier membranes and the substan-
tial overlap between their substrates and many drugs. Unfortunately, because
of our limited understanding of transporters, most prediction programs do
not have a mechanism to incorporate the effect of active transport. However,
interest in these transporters has resulted in a relatively large amount of in
vitro data, which in turn have enabled the generation of pharmacophore and
QSAR models for many of them. These models have assisted in the under-
standing of the complex effects of transporters on drug disposition, including
absorption, distribution, and excretion. Their incorporation into current mod-



eling programs would also result in more accurate prediction of drug disposi-
tion behavior. Readers are referred to a recent review for discussions of in
silico strategies in modeling transporters [41].

20.6.1 P-gp

P-glycoprotein (P-gp) is an ATP-dependent effl ux transporter that transports
a broad range of substrates out of the cell. It affects drug disposition by reduc-
ing absorption and enhancing renal and hepatic excretion [42]. For example,
P-gp is known to limit the intestinal absorption of the anticancer drug pacli-
taxel [43] and restricts the CNS penetration of human immunodefi ciency
virus (HIV) protease inhibitors [44]. It is also responsible for multidrug resis-
tance in cancer chemotherapy. Because of its signifi cance in drug disposition
and effective cancer treatment, P-gp attracted numerous efforts and has
become the most extensively studied transporter, with abundant experimental
data [42].

Ekins and colleagues generated fi ve computational pharmacophore models
to predict the inhibition of P-gp from in vitro data on a diverse set of inhibi-
tors with several cell systems, including inhibition of digoxin transport and
verapamil binding in Caco-2 cells; vinblastine and calcein accumulation in
P-gp-expressing LLC-PK1 (L-MDR1) cells; and vinblastine binding in vesi-
cles derived from CEM/VLB100 cells [45, 46]. By comparing and merging all
P-gp pharmacophore models, common areas of identical chemical features
such as hydrophobes, hydrogen bond acceptors, and ring aromatic features as
well as their geometric arrangement were identifi ed to be the substrate
requirements for P-gp. Similar transport requirements were reiterated in
other works [47, 48]. More recently Cianchetta and colleagues combined
alignment-independent 3D descriptors and physicochemical descriptors to
model inhibition of calcein accumulation in Caco-2 cells [49]. Using a diverse
set of 129 compounds, the authors derived a robust QSAR model that revealed
two hydrophobic features, two hydrogen bond acceptors, and the molecular
dimension to be essential determinants of P-gp-mediated transport. These
identifi ed transport requirements not only to help screen compounds with
potential effl ux related bioavailability problems, but also to assist the identi-
fi cation of novel P-gp inhibitors, which when coadministered with target drugs
would optimize their pharmacokinetic profi le by increasing bioavailability. In
fact, a recent pharmacophore-based database screening has proposed 28
novel P-gp inhibitors from the Derwent World Drug Index [50]. Our own
Catalyst pharmacophore searches of databases have also guided the identifi –
cation of several currently prescribed drugs that are P-gp inhibitors (µM),
which was previously unknown (Fig. 20.2, manuscript in preparation).

20.6.2 BCRP

Breast cancer resistance protein (BCRP) is another ATP-dependent effl ux
transporter that confers resistance to a variety of anticancer agents, including



Figure 20.2 Pharmacophore models for P-gp inhibition. A. P-gp inhibition pharma-
cophore aligned with the potent inhibitor LY335979. B. P-gp substrate pharmaco-
phore aligned with verapamil. C. P-gp inhibition pharmacophore 2 aligned with
LY335979. Green indicates H-bond acceptor feature, and cyan indicates hydrophobic
feature. See color plate.

anthracyclines and mitoxantrone [51]. In addition to a high level of expression
in hematological malignancies and solid tumors, BCRP is also expressed in
intestine, liver, and brain, thus implicating its intricate role in drug disposition
behavior. Recently, Zhang and colleagues generated a BCRP 3D-QSAR
model by analyzing structure and activity of 25 fl avonoid analogs [52]. The
model emphasizes very specifi c structural feature requirements for BCRP
such as the presence of a 2,3-double bond in ring C and hydroxylation at
position 5. Because the model is only based on a set of closely related struc-
tures instead of a diverse set, it should be applied with caution. Satisfying the
transport model would render a compound susceptible to BCRP, but not
fi tting into the model does not necessarily exclude the candidate from BCRP
transport. In fact, this caveat should be considered for all predictive in silico
models, because no model can cover all possible chemical space.

20.6.3 Nucleoside Transporters

Nucleoside transporters transport both naturally occurring nucleosides and
synthetic nucleoside analogs that are used as anticancer drugs (e.g., cladrib-
ine) and antiviral drugs (e.g., zalcitabine). There are different types of nucle-
oside transporters, including concentrative nucleoside transporters (CNT1,
CNT2, CNT3) and equilibrative nucleoside transporters (ENT1, ENT2),
each having different substrate specifi cities. The broad-affi nity, low-selective
ENTs are ubiquitously located, whereas the high-affi nity, selective CNTs are
mainly located in epithelia of intestine, kidney, liver, and brain [53], indicat-
ing their involvement in drug absorption, distribution, and excretion. The
fi rst 3D-QSAR model for nucleoside transporters was generated back in
1990 [54]. It is an oversimplifi ed general model limited by the scarce experi-
mental data at that time. A more comprehensive study generated distinctive
models for CNT1, CNT2, and ENT1 with both pharmacophore and 3D-
QSAR modeling techniques [55]. All models show the common features



required for nucleoside transporter-mediated transport: two hydrophobic
features and one hydrogen bond acceptor on the pentose ring. The individual
models also reveal the subtle characteristic requirements for each specifi c
transporter. The modeling results also support the previous observation that
CNT2 is the most selective transporter whereas ENT1 has the broadest
inhibitor specifi city. More recently, we performed the same analyses and
generated pharmacophore and 3D-QSAR models for CNT3 by assessing the
transport activity of 33 nucleoside analogs [55a]. These studies represent a
comprehensive evaluation of transport requirements of all three types of

20.6.4 hPEPT1

The human peptide transporter (hPEPT1) is a low-affi nity high-capacity oli-
gopeptide transport system that transports a diverse range of substrates
including β-lactam antibiotics [56] and angiotensin-converting enzyme (ACE)
inhibitors [57]. It is mainly expressed in intestine and kidney, affecting drug
absorption and excretion. A pharmacophore model based on three high-
affi nity substrates (Gly-Sar, bestatin, and enalapril) recognized two hydro-
phobic features, one hydrogen bond donor, one hydrogen bond acceptor, and
one negative ionizable feature to be hPEPT1 transport requirements [58].
This pharmacophore model was subsequently applied to screen the CMC
database with over 8000 druglike molecules. The antidiabetic repaglinide and
HMG-CoA reductase inhibitor fl uvastatin were suggested by the model and
later verifi ed to inhibit hPEPT1 with submillimolar potency [58]. This work
demonstrated the potential of applying in silico models in high-throughput
database screening.

20.6.5 ASBT

The human apical sodium-dependent bile acid transporter (ASBT) is a high-
effi cacy, high-capacity transporter expressed on the apical membrane of intes-
tinal epithelial cells and cholangiocytes. It assists absorption of bile acids and
their analogs, thus providing an additional intestinal target for improving
drug absorption. Baringhaus and colleagues developed a pharmacophore
model based on a training set of 17 chemically diverse inhibitors of ASBT
[59]. The model revealed ASBT transport requirements as one hydrogen
bond donor, one hydrogen bond acceptor, one negative charge, and three
hydrophobic centers. These requirements are in good agreement with a previ-
ous 3D-QSAR model derived from the structure and activity of 30 ASBT
inhibitors and substrates [60].

20.6.6 OCT

The organic cation transporters (OCTs) facilitate the uptake of many cationic
drugs across different barrier membranes from kidney, liver, and intestine



epithelia. A broad range of drugs or their metabolites fall into the chemical
class of organic cation (carrying a net positive charge at physiological pH)
including antiarrhythmics, β-adrenoreceptor blocking agents, antihistamines,
antiviral agents, and skeletal muscle-relaxing agents [61]. Three OCTs have
been cloned from different species, OCT1, OCT2, and OCT3. A human
OCT1 pharmacophore model was developed by analyzing the extent of inhi-
bition of TEA uptake in HeLa cells of 22 diverse molecules. The model sug-
gests the transport requirements of human OCT1 as three hydrophobic
features and one positive ionizable feature [62]. Molecular determinants of
substrate binding to human OCT2 and rabbit OCT2 were recently reported
[63]. Both 2D- and 3D-QSAR analyses were performed to identify and dis-
criminate the binding requirements of the two orthologs. The models showed
the same chemical features, highlighting their similarities. However, the ori-
entation of a critical hydrogen bonding feature set the two orthologs apart.
This work illustrates the sensitivity of in silico modeling in discriminating
similar transporters.

20.6.7 OATP

Organic anion transporting polypeptides (OATPs) infl uence the plasma con-
centration of many drugs by actively transporting them across a diverse range
of tissue membranes such as liver, intestine, lung, and brain [64]. Because of
their broad substrate specifi city, OATPs transport not only organic anionic
drugs, as originally thought, but also organic cationic drugs. Currently 11
human OATPs have been identifi ed, and the substrate binding requirements
of the best-studied OATP1B1 were successfully modeled with the metaphar-
macophore approach recently [65]. Through assessing a training set of 18
diverse molecules, the metapharmacophore model identifi ed three hydropho-
bic features fl anked by two hydrogen bond acceptor features to be the essen-
tial requirement for OATP1B1 transport. Similar requirements were derived
from another 3D-QSAR study based on rat Oatp1a5 [66].

20.6.8 BBB-Choline Transporter

The BBB-choline transporter is a native nutrient transporter that transports
choline, a charged cation, across the BBB into the CNS [67]. Its active trans-
port assists the BBB penetration of cholinelike compounds, and understand-
ing its structural requirements should afford a more accurate prediction of
BBB permeation. Even though the BBB-choline transporter has not been
cloned, Geldenhuys and colleagues applied a combination of empirical and
theoretical methodologies to study its binding requirements [68]. The 3D-
QSAR models were built with empirical Ki data obtained from in situ rat
brain perfusion experiments with a structurally diverse set of compounds.
Three hydrophobic interactions and one hydrogen bonding interaction sur-
rounding the positively charged ammonium moiety were identifi ed to be
important for BBB-choline transporter recognition. Even though the model



statistical signifi cance is not optimal (q2 < 0.5), it does provide a useful esti-
mation of BBB-choline transporter binding requirements. More accurate in
silico models could be generated once higher-quality data from the cloned
BBB-choline transporter are available.


Two years ago, several reviews [e.g., 2, 13] pointed out that data quality is the
most limiting factor in ADMET modeling. We believe that data quality is still
the weakest link, thereby effectively limiting the practical application of
ADMET models. The major recent advancement in ADMET modeling is in
elucidating the role and successful modeling of various transporters [45, 46,
48, 50, 52, 55, 58–60, 62, 63, 65, 66, 68, 69]. Incorporation of the infl uence of
these transporters into current models is an ongoing task in ADMET model-
ing. Some commercial programs have already implemented the capability of
modeling active transport, such as the recent versions of GastroPlus (Simula-
tions Plus, Lancaster, CA), PK-Sim (Bayer Technology Services, Germany),
and ADME/Tox WEB (Pharma Algorithms, Toronto, ON, Canada). A suc-
cessful implementation of active transport as a fi lter is exemplifi ed in the
ADME/Tox WEB absorption prediction program [70]. Compounds are fi rst
screened against pharmacophore models of different active transporters. The
compound that fi ts these models is removed from further predictions, which
is based solely on physicochemical properties.

Importantly, the currently available transporter models only cover a small
fraction of all transporters involved in drug disposition. Other than incorpo-
rating current stand-alone transporter models into systemic models to directly
predict drug pharmacokinetic properties, continued efforts are still needed
to investigate other transporters such as MRP, BCRP, NTCP, and OAT, to
get a more complete understanding of the drug pharmacokinetic profi le.

Not all pharmaceutical companies can afford the resources to generate
their own in-house modeling programs, so the commercially available in silico
modeling suites have become an attractive option. However, this leads to a
potential problem: The chemical space that these commercial packages are
developed from might not be directly related to the company’s chemical
scope. In silico models are most predictive when applied in the same chemical
space as the training compounds. As a result, a decreased predictive power
is to be expected when the model is applied to a different chemical space. The
fact that the majority of these programs do not offer capabilities to customize
parameters aggravates the above-mentioned problem. In answer to this, some
modeling programs such as Algorithm Builder (Pharma Algorithms, Toronto,
ON, Canada) are offering fl exibility for customers to generate their in-house
models with their own training set and the statistical algorithm of their choice.
Additionally, we should expect more mechanism-based modeling algorithms
that are easy to understand and implement owing to a more detailed under-
standing of underlying mechanisms for different aspects of drug disposition.



These trends will accelerate the shift of model building from computational
scientists to experimental scientists.

As discussed above, all ADMET aspects are dependent on each other and
should all be considered when making predictions. Integrated analysis of dif-
ferent aspects of drug pharmacokinetic profi les is yet another future trend.
Ultimately, drug ADMET properties should be predicted based on an inte-
gration of a compilation of in silico models refl ecting different aspects of the


1. Ekins S, Nikolsky Y and Nikolskaya T. Techniques: Application of systems
biology to absorption, distribution, metabolism, excretion, and toxicity. Trends
Pharmacol Sci 2005;26:202–9.

2. van de Waterbeemd H and Gifford E. ADMET in silico modelling: towards pre-
diction paradise? Nat Rev Drug Discov 2003;2:192–204.

3. Martin YC, Bures MG, Danaher EA, DeLazzer J, Lico I and Pavlik PA. A fast
new approach to pharmacophore mapping and its application to dopaminergic
and benzodiazepine agonists. J Comput Aided Mol Des 1993;7:83–102.

4. Jones G, Willett P and Glen RC. A genetic algorithm for fl exible molecular
overlay and pharmacophore elucidation. J Comput Aided Mol Des 1995;9:

5. Clement OO and Mehl AT. HipHop: pharmacophore based on multiple common-
feature alignments. In: Guner OF, editor, Pharmacophore perception, develop-
ment, and use in drug design. San Diego: IUL, 2000. pp. 69–84.

6. Patel Y, Gillet VJ, Bravi G and Leach AR. A comparison of the pharmacophore
identifi cation programs: Catalyst, DISCO and GASP. J Comput Aided Mol Des

7. Schneidman-Duhovny D, Nussinov R and Wolfson HJ. Predicting molecular
interactions in silico: II. Protein-protein and protein-drug docking. Curr Med
Chem 2004;11:91–107.

8. Cruciani G, Pastor M and Guba W. VolSurf: a new tool for the pharmacokinetic
optimization of lead compounds. Eur J Pharm Sci 2000;11 Suppl 2:S29–39.

9. Eros D, Keri G, Kovesdi I, Szantai-Kis C, Meszaros G and Orfi L. Comparison
of predictive ability of water solubility QSPR models generated by MLR, PLS
and ANN methods. Mini Rev Med Chem 2004;4:167–77.

10. Yang G, Ran Y and Yalkowsky SH. Prediction of the aqueous solubility: com-
parison of the general solubility equation and the method using an amended sol-
vation energy relationship. J Pharm Sci 2002;91:517–33.

11. Butina D and Gola JM. Modeling aqueous solubility. J Chem Inf Comput Sci

12. Delaney JS. ESOL: estimating aqueous solubility directly from molecular struc-
ture. J Chem Inf Comput Sci 2004;44:1000–5.

13. Lombardo F, Gifford E and Shalaeva MY. In silico ADME prediction: data,
models, facts and myths. Mini Rev Med Chem 2003;3:861–75.



14. Refsgaard HH, Jensen BF, Brockhoff PB, Padkjaer SB, Guldbrandt M and Chris-
tensen MS. In silico prediction of membrane permeability from calculated molec-
ular parameters. J Med Chem 2005;48:805–11.

15. Fujikawa M, Ano R, Nakao K, Shimizu R and Akamatsu M. Relationships
between structure and high-throughput screening permeability of diverse drugs
with artifi cial membranes: application to prediction of Caco-2 cell permeability.
Bioorg Med Chem 2005;13:4721–32.

16. Artursson P and Karlsson J. Correlation between oral drug absorption in humans
and apparent drug permeability coeffi cients in human intestinal epithelial
(Caco-2) cells. Biochem Biophys Res Commun 1991;175:880–5.

17. Irvine JD, Takahashi L, Lockhart K, Cheong J, Tolan JW, Selick HE, et al.
MDCK (Madin-Darby canine kidney) cells: a tool for membrane permeability
screening. J Pharm Sci 1999;88:28–33.

18. Malkia A, Murtomaki L, Urtti A and Kontturi K. Drug permeation in biomem-
branes: in vitro and in silico prediction and infl uence of physicochemical proper-
ties. Eur J Pharm Sci 2004;23:13–47.

19. Agoram B, Woltosz WS and Bolger MB. Predicting the impact of physiological
and biochemical processes on oral drug bioavailability. Adv Drug Deliv Rev
2001;50 Suppl 1:S41–67.

20. Yu LX, Lipka E, Crison JR and Amidon GL. Transport approaches to the bio-
pharmaceutical design of oral drug delivery systems: prediction of intestinal
absorption. Adv Drug Deliv Rev 1996;19:359–76.

21. Lombardo F, Obach RS, Shalaeva MY and Gao F. Prediction of volume of dis-
tribution values in humans for neutral and basic drugs using physicochemical
measurements and plasma protein binding data. J Med Chem 2002;45:2867–76.

22. Lombardo F, Obach RS, Shalaeva MY and Gao F. Prediction of human volume
of distribution values for neutral and basic drugs. 2. Extended data set and leave-
class-out statistics. J Med Chem 2004;47:1242–50.

23. Colmenarejo G, Alvarez-Pedraglio A and Lavandera JL. Cheminformatic models
to predict binding affi nities to human serum albumin. J Med Chem 2001;44:

24. Kratochwil NA, Huber W, Muller F, Kansy M and Gerber PR. Predicting plasma
protein binding of drugs: a new approach. Biochem Pharmacol 2002;64:

25. Lobell M and Sivarajah V. In silico prediction of aqueous solubility, human
plasma protein binding and volume of distribution of compounds from calculated
pKa and AlogP98 values. Mol Divers 2003;7:69–87.

26. Mager DE and Jusko WJ. Quantitative structure-pharmacokinetic/pharmacody-
namic relationships of corticosteroids in man. J Pharm Sci 2002;91:2441–51.

27. Yamazaki K and Kanaoka M. Computational prediction of the plasma protein-
binding percent of diverse pharmaceutical compounds. J Pharm Sci 2004;93:

28. Cabrera MA, Bermejo M, Perez M and Ramos R. TOPS-MODE approach for
the prediction of blood-brain barrier permeation. J Pharm Sci 2004;93:1701–17.

29. Ecker GF and Noe CR. In silico prediction models for blood-brain barrier per-
meation. Curr Med Chem 2004;11:1617–28.



30. Li H, Yap CW, Ung CY, Xue Y, Cao ZW and Chen YZ., Effect of selection of
molecular descriptors on the prediction of blood-brain barrier penetrating and
nonpenetrating agents by statistical learning methods. J Chem Inf Model

31. Narayanan R and Gunturi SB. In silico ADME modelling: prediction models for
blood-brain barrier permeation using a systematic variable selection method.
Bioorg Med Chem 2005;13:3017–28.

32. Subramanian G and Kitchen DB. Computational models to predict blood-brain
barrier permeation and CNS activity. J Comput Aided Mol Des 2003;17:643–

33. Pardridge WM, Log(BB). PS products and in silico models of drug brain penetra-
tion. Drug Discov Today 2004;9:392–3.

34. Abraham MH. The factors that infl uence permeation across the blood-brain
barrier. Eur J Med Chem 2004;39:235–40.

35. Liu X, Tu M, Kelly RS, Chen C and Smith BJ. Development of a computational
approach to predict blood-brain barrier permeability. Drug Metab Dispos

36. Goodwin JT and Clark DE., In silico predictions of blood-brain barrier penetra-
tion: considerations to “keep in mind”. J Pharmacol Exp Ther 2005.

37. Taylor EM. The impact of effl ux transporters in the brain on the development of
drugs for CNS disorders. Clin Pharmacokinet 2002;41:81–92.

38. Toutain PL and Bousquet-Melou A. Plasma clearance. J Vet Pharmacol Ther

39. Ito K and Houston JB. Prediction of human drug clearance from in vitro and
preclinical data using physiologically based and empirical approaches. Pharm Res

40. Sasaki M, Suzuki H, Aoki J, Ito K, Meier PJ and Sugiyama Y. Prediction of in
vivo biliary clearance from the in vitro transcellular transport of organic anions
across a double-transfected Madin-Darby canine kidney II monolayer expressing
both rat organic anion transporting polypeptide 4 and multidrug resistance asso-
ciated protein 2. Mol Pharmacol 2004;66:450–9.

41. Chang C, Ray A and Swaan P. In silico strategies for modeling membrane trans-
porter function. Drug Discov Today 2005;10:663–71.

42. Lin JH and Yamazaki M. Role of P-glycoprotein in pharmacokinetics: clinical
implications. Clin Pharmacokinet 2003;42:59–98.

43. Sparreboom A, van Asperen J, Mayer U, Schinkel AH, Smit JW, Meijer DK, et
al. Limited oral bioavailability and active epithelial excretion of paclitaxel (Taxol)
caused by P-glycoprotein in the intestine. Proc Natl Acad Sci USA

44. Polli JW, Jarrett JL, Studenberg SD, Humphreys JE, Dennis SW, Brouwer KR,
et al. Role of P-glycoprotein on the CNS disposition of amprenavir (141W94), an
HIV protease inhibitor. Pharm Res 1999;16:1206–12.

45. Ekins S, Kim RB, Leake BF, Dantzig AH, Schuetz E, Lan LB, et al. Application
of three dimensional quantitative structure-activity relationships of P-
glycoprotein inhibitors and substrates. Mol Pharmacol 2002;61:974–981.



46. Ekins S, Kim RB, Leake BF, Dantzig AH, Schuetz EG, Lan LB, et al. Three-
dimensional quantitative structure-activity relationships of inhibitors of P-
glycoprotein. Mol Pharmacol 2002;61:964–73.

47. Pajeva IK and Wiese M. Pharmacophore model of drugs involved in P-glycoprotein
multidrug resistance: explanation of structural variety (hypothesis). J Med Chem

48. Yates CR, Chang C, Kearbey JD, Yasuda K, Schuetz EG, Miller DD, et al. Struc-
tural determinants of P-glycoprotein-mediated transport of glucocorticoids.
Pharm Res 2003;20:1794–803.

49. Cianchetta G, Singleton RW, Zhang M, Wildgoose M, Giesing D, Fravolini A, et
al. A pharmacophore hypothesis for P-glycoprotein substrate recognition using
GRIND-based 3D-QSAR. J Med Chem 2005;48:2927–35.

50. Langer T, Eder M, Hoffmann RD, Chiba P and Ecker GF. Lead identifi cation
for modulators of multidrug resistance based on in silico screening with a phar-
macophoric feature model. Arch Pharm (Weinheim) 2004;337:317–27.

51. Mao Q and Unadkat JD. Role of the breast cancer resistance protein (ABCG2)
in drug transport. AAPS J 2005;7:E118–33.

52. Zhang S, Yang X, Coburn RA and Morris ME. Structure activity relationships
and quantitative structure activity relationships for the fl avonoid-mediated
inhibition of breast cancer resistance protein. Biochem Pharmacol

53. Aymerich I, Dufl ot S, Fernandez-Veledo S, Guillen-Gomez E, Huber-Ruano I,
Casado FJ, et al. The concentrative nucleoside transporter family (SLC28): new
roles beyond salvage? Biochem Soc Trans 2005;33:216–9.

54. Viswanadhan VN, Ghose AK and Weinstein JN. Mapping the binding site of the
nucleoside transporter protein: a 3D-OSAR study. Biochim Biophys Acta

55. Chang C, Swaan PW, Ngo LY, Lum PY, Patil SD and Unadkat JD. Molecular
requirements of the human nucleoside transporters hCNT1, hCNT2, and hENT1.
Mol Pharmacol 2004;65:558–70.

55a. Hu H, Endres CJ, Chang C, Umapathy NS, Lee EW, Fei YJ, et al. Electrophysi-
ological Characterization and Modeling of the Structure Activity Relationship of
the Human Concentrative Nucleoside Transporter 3 (hCNT3). Mol Pharmacol
2006; in press.

56. Snyder NJ, Tabas LB, Berry DM, Duckworth DC, Spry DO and Dantzig AH.
Structure-activity relationship of carbacephalosporins and cephalosporins: anti-
bacterial activity and interaction with the intestinal proton-dependent dipeptide
transport carrier of Caco-2 cells. Antimicrob Agents Chemother 1997;41:1649–57.

57. Leibach FH and Ganapathy V. Peptide transporters in the intestine and the
kidney. Annu Rev Nutr 1996;16:99–119.

58. Ekins S, Johnston JS, Bahadduri P, D’Souza VM, Ray A, Chang C, et al. In vitro
and pharmacophore-based discovery of novel hPEPT1 inhibitors. Pharm Res

59. Baringhaus KH, Matter H, Stengelin S and Kramer W. Substrate specifi city of
the ileal and the hepatic Na+ /bile acid cotransporters of the rabbit. II. A reliable



3D QSAR pharmacophore model for the ileal Na+ /bile acid cotransporter.
J Lipid Res 1999;40:2158–68.

60. Swaan PW, Szoka FC, Jr. and Oie S. Molecular modeling of the intestinal bile
acid carrier: a comparative molecular fi eld analysis study. J Comput Aided Mol
Des 1997;11:581–8.

61. Wright SH and Dantzler WH. Molecular and cellular physiology of renal organic
cation and anion transport. Physiol Rev 2004;84:987–1049.

62. Bednarczyk D, Ekins S, Wikel JH and Wright SH. Infl uence of molecular struc-
ture on substrate binding to the human organic cation transporter, hOCT1. Mol
Pharmacol 2003;63:489–98.

63. Suhre WM, Ekins S, Chang C, Swaan PW and Wright SH. Molecular determi-
nants of substrate/inhibitor binding to the human and rabbit renal organic cation
transporters hOCT2 and rbOCT2. Mol Pharmacol 2005;67:1067–77.

64. Tamai I, Nezu J, Uchino H, Sai Y, Oku A, Shimane M, et al. Molecular identifi ca-
tion and characterization of novel members of the human organic anion trans-
porter (OATP) family. Biochem Biophys Res Commun 2000;273:251–60.

65. Chang C, Pang KS, Swaan PW, Ekins S. Comparative pharmacophore modeling
of organic anion transporting polypeptides: a meta-analysis of rat Oatp1a1 and
human OATP1B1. J Pharmacol Exp Ther 2005; 314:533–41.

66. Yarim M, Moro S, Huber R, Meier PJ, Kaseda C, Kashima T, et al. Application
of QSAR analysis to organic anion transporting polypeptide 1a5 (Oatp1a5) sub-
strates. Bioorg Med Chem 2005;13:463–71.

67. Allen DD and Smith QR. Characterization of the blood-brain barrier choline
transporter using the in situ rat brain perfusion technique. J Neurochem

68. Geldenhuys WJ, Lockman PR, McAfee JH, Fitzpatrick KT, Van der Schyf CJ
and Allen DD. Molecular modeling studies on the active binding site of
the blood-brain barrier choline transporter. Bioorg Med Chem Lett

69. Zhang EY, Phelps MA, Cheng C, Ekins S and Swaan PW. Modeling of active
transport systems. Adv Drug Deliv Rev 2002;54:329–54.

70. Zmuidinavicius D, Didziapetris R, Japertas P, Avdeef A and Petrauskas A. Clas-
sifi cation structure-activity relations (C-SAR) in prediction of human intestinal
absorption. J Pharm Sci 2003;92:621–33.