Agoni 2020

Download as pdf or txt
Download as pdf or txt
You are on page 1of 11

Journal of Molecular Modeling (2020) 26:120

https://doi.org/10.1007/s00894-020-04385-6

REVIEW

Druggability and drug-likeness concepts in drug design: are


biomodelling and predictive tools having their say?
Clement Agoni 1 & Fisayo A. Olotu 1 & Pritika Ramharack 1 & Mahmoud E. Soliman 1

Received: 8 October 2019 / Accepted: 22 April 2020


# Springer-Verlag GmbH Germany, part of Springer Nature 2020

Abstract
The drug discovery process typically involves target identification and design of suitable drug molecules against these targets.
Despite decades of experimental investigations in the drug discovery domain, about 96% overall failure rate has been recorded in
drug development due to the “undruggability” of various identified disease targets, in addition to other challenges. Likewise, the
high attrition rate of drug candidates in the drug discovery process has also become an enormous challenge for the pharmaceutical
industry. To alleviate this negative outlook, new trends in drug discovery have emerged. By drifting away from experimental
research methods, computational tools and big data are becoming valuable in the prediction of biological target druggability and
the drug-likeness of potential therapeutic agents. These tools have proven to be useful in saving time and reducing research costs.
As with any emerging technique, however, controversial opinions have been presented regarding the validation of predictive
computational tools. To address the challenges associated with these varying opinions, this review attempts to highlight the
principles of druggability and drug-likeness and their recent advancements in the drug discovery field. Herein, we present the
different computational tools and their reliability of predictive analysis in the drug discovery domain. We believe that this report
would serve as a comprehensive guide towards computational-oriented drug discovery research.

Keywords Druggability . Drug-likeness . Computer-aided drug design . Drug discovery

Introduction overall failure rate has been recorded in diverse drug discov-
ery projects due to the “undruggability” of various identified
The design and development of several therapeutic agents protein targets, amidst other challenges [8–10]. The drug dis-
centered on their ability to interact with crucial biomolecules covery process is usually a time-consuming process and sub-
involved in disease progression. These biological molecules stantial financial investment; as such, a 96% failure of the
majorly include proteins and enzymes that drive pathogenic process due to target undruggability and other challenges ne-
pathways and could represent therapeutic avenues for halting cessitates an understanding of challenges associated with tar-
undesirable pathogenic development such as in cancer, TB, get’s undruggability.
and neurodegenerative diseases, some of which have been As essential as lead identification and lead optimization
termed “undruggable” [1–4]. Of the 10% of the human ge- techniques are in drug discovery, the concept of protein
nome representing druggable targets for therapeutic agents, druggability remains vital, usually descriptive of members of
only half of those are relevant to disease [5–7]. Over the years, expressed protein cohort (proteome) that can selectively be
a significant advancement has been in the experimental and bound and modulated by high-affinity ligands with innate
computational techniques employed in the drug discovery drug-likeness [11, 12]. Studies have shown that among the
process; nonetheless, a remarkable failure of about 96% disparate protein families that constitute the human proteome,
about 3000 have been targeted by small synthetic/organic
drug molecules with regard to therapeutic intervention in dis-
* Mahmoud E. Soliman eases [13–16]. As such, there remains a large pool of other
soliman@ukzn.ac.za; http://soliman.ukzn.ac.za proteins that belong to families that are not targetable or
1 “uninhibitable” by small drug molecules.
Molecular Bio-Computation & Drug Design Lab, School of Health
Sciences, University of KwaZulu-Natal, Westville, Durban 4000, The concept of druggability has also expanded to encom-
South Africa pass the “druggable genome,” which describes encoding
120 Page 2 of 11 J Mol Model (2020) 26:120

genes for proteins that are not only physiologically relevant mass, hydrophobicity, and polarity when compared to the nar-
but whose activity can also be regulated by high-affinity li- row range distribution of approved drugs [30, 31]. Regardless,
gands [11, 14, 16]. It is also important to mention that drug-like compounds hold no apparent structural similarity to
druggability as a term is more suitable to describe protein any approved drug. Although high rates of attrition have
targets than DNA and RNA sequences, even though they shown in drug discovery, the preferential selection of com-
may also serve as active sites for drug molecules [12, 17]. pounds with innate drug-likeness increases the likelihood of
An integral component of drug development in recent years surviving this occurrence [32].
has, therefore, been target validation. Existing conventional A critical understanding of the concepts of target
processes of target validation techniques are known to inves- druggability coupled with the drug-likeness of a chemical
tigate the interplay between alteration in protein biological molecule could, therefore, further potentiate the success rate
functionality and corresponding therapeutic impacts. Thus, of novel chemical molecules towards becoming viable thera-
druggability incorporates a structure-based approach to eval- peutic options in disease intervention.
uate tendencies that small drug-like molecules possess innate Therefore, in this review, we discuss the underlying princi-
potencies to bind and modulate protein activities [11, 13]. ples of both druggability and drug-likeness as it relates to the
Consequently, several approaches have emerged over the drug discovery process. Also, we list the various methods that
years to investigate the druggability of potential protein targets have been employed over the years to predict, determine, and
as succinctly discussed by Egner and Hillig [18]. achieve the druggability of a novel protein target as previously
Approaches employed to determine the druggability of a studied. We extend this methodical review to cover evolution-
protein have been categorized based on (1) those that employ ary advances made in determining the drug-likeness of a com-
empirical analyses of established associations between drugs pound coupled with relevant experimental and computational
and proteins and (2) structural analyses of target proteins [13, techniques aimed at enhancing the drug-likeness of a nondrug
19–23]. Moreover, when the protein in question is a member hit compound.
of a well-studied family with closely related homologs from
which drug association details can be retrieved, the former
method is more suitably applied [15, 16]. On the other hand, Assessing druggability of biological targets
where there is no prior knowledge about the potential target or
in cases where the target protein belongs to families for which Advancements in the computational domain have paved a
no drug candidates are known, the latter method that employs path towards algorithms that assess the druggability of a target
structure-based predictive models is most appropriate [13]. and store any information in large virtual libraries. We now
A notable and pioneering contribution to the development have access to a plethora of data, including protein and gene
of the structural predictive model was carried out by Hajduk expression profiles, protein-protein interactions, and gene reg-
et al., wherein they measured druggability using hit rates of ulatory networks, thus making it possible to build combined
NMR-based fragment screening for a diverse set of 23 pro- predictive models. Initial algorithms used to assess
teins. This model incorporates a simple regression analysis druggability were designed to reproduce nuclear magnetic
that defined descriptors of the binding pocket such as the resonance (NMR) fragment screening hit rates [21, 33, 34].
surface area, roughness, polar/apolar contact area, and the This approach, however, leads to proprietary data that was
number of charged residues [21, 24]. Their analysis generated expensive to obtain and difficult to access publicly. Another
a predictive “druggability score,” which defines the logarithm method of assessing druggability was reported by Cheng et al.
of the predicted rate of NMR hit. Over time, a variety of in which the maximum achievable affinity for a binding site
methods have been developed to improve the prediction of based on the structure of the pocket was estimated [17].
protein druggability [11, 13, 19, 23, 25–27], some of which Likewise, a separate report by Sheridan et al. also described
form an integral part of this review. an equation that could approximate the “bindability” of a giv-
Drug-likeness is a significant consideration in the early en binding pocket on a protein, based on similar functions and
stages of the drug discovery process when selecting com- some particular descriptors as methods for assessing
pounds as it helps to optimize pharmacokinetics and pharma- druggability [19]. Experimental druggability assessments in-
ceutical attributes. These include chemical stability, solubility, volved soaking protein crystals in various solvents and iden-
bioavailability, and distribution profile [12]. Drug-like com- tifying binding sites based on X-ray diffraction data. Although
pounds also describe chemical molecules that possess func- this method was the most reliable and validated technique, the
tional groups and or have physical properties consistent with crystallized protein was required [34, 35].
the majority of known drugs and hence can be inferred as Various computational tools have been developed over the
compounds that might be active biologically or might show year to assess the druggability of a target based on the availability
therapeutic potential [28, 29]. Thus, these compounds fall of a protein’s 3D structure. Some of these tools are based on
within critical physicochemical thresholds such as molecular algorithms that consider structural features of a protein, including
J Mol Model (2020) 26:120 Page 3 of 11 120

Fig. 1 Methods for assessing the druggability of potential disease targets

surface polarity, surface hydrophobicity, and pocket dimensions. Sequenced-based druggability assessment methods
The algorithms used are roughly classified into geometric and
energetic approaches [36, 37]. In the geometric approach, the The basic unit of proteins, amino acids, which determine the
shape of the protein is investigated to identify any voids between function of a protein, has also been employed in predicting the
the protein’s atoms. The energetic approach uses interactive mo- druggability of protein drug targets. As such, sequence-based
lecular energies, which are mapped onto the 3D structure of the druggability assessment methods analyze sequence align-
protein through molecular dynamics simulations. The binding ments of protein targets, identifying possible conserved re-
pockets are then identified based on the interaction energy pro- gions that may be deemed predictive binding pockets [34].
files that are generated [34, 37]. The downfall of the approaches Assessment of potential target sequence also provides insights
mentioned above is that a 3D structure of the protein is needed, into essential physicochemical properties like solvent accessi-
the process may be slow, and the predictive pockets cannot be bility, hydrophobicity, charge, and polarity, among other prop-
validated until an experimental analysis is performed. Because erties [41]. Sequence-based methods have also been wildly
not all druggable targets are proteins, there are a few databases employed in machine-learning algorithms in assessing protein
that identify other druggable targets, notably Therapeutic Target target druggability [41]. Even though a knowledge of the tar-
Database [38], LncRNA2Target [39], and MiRBase [40]. get sequence could aid in the prediction of functional domains
Four primary methods are currently used for assessing on the target, the target sequence alone is insufficient in pro-
druggability; these include the precedence-based, ligand- viding an overall structural and functional insight of a protein
based, structure-based, and sequence-based methods, as target. Thus, the knowledge of target sequence alone is unable
highlighted in Fig. 1 [11]. Although each method comes with to provide details on the accessibility of these domains to
its benefits, numerous caveats need to be considered and either gene expression level, a drug molecule, or its relevance
accounted for when identifying an ideal target. One of the in the interactome [41].
most important considerations that are frequently overlooked
is that compounds do not always target a protein. Other Structure-based druggability assessment
existing biological targets include nucleic acids, other small
molecules, protein-protein interactions, and nucleic acid com- Structure-based druggability assessment methods rely on a
plexes such as ribosomes [35]. knowledge of the binding pocket located on the biological
120 Page 4 of 11 J Mol Model (2020) 26:120

target to which a ligand can bind [34]. Thus, for a small drug- 1364 approved biologics (proteins, peptides, vaccines, and
like molecule to bind a protein, it requires a binding site that allergenics), 130 nutraceuticals, and over 6347 experimental
contains complementary characteristics, including an appro- (discovery-phase) drugs.
priate size that can accommodate a molecule and a deep hy-
drophobic pocket that can bury the drug within the protein.
Large exposed polar binding sites are considered less
druggable than smaller hydrophobic pockets [34, 42]. The Are predictive druggability assessment
structure-based assessment methods generally involve an ini- methods reliable?
tial identification of possible binding pockets followed by
sorting the identified binding pockets based on the physico- While investigations of most disease targets have potential
chemical properties of the pocket. Sorted binding pockets are inhibitors, there is minimal experimental data associated with
then compared to a reference set of known biological targets to the research. This observation has led to almost all scientific
access their degrees of difficulty or druggability [43]. domains to turn to machine learning and predictive computa-
Numerous software and online platforms have been devel- tional tools. Not only do these methods save laborious hours
oped to aid in the accurate identification of binding pockets, in the wet-laboratory setting but it also saves elaborate
notably DOGSiteScorer [26], Metapocket [44], PockDrug amounts of money that would have been spent on unnecessar-
Server [45], SiteMap [23], Sitehound [46], and Open Targets ily large stocks of reagents. Therefore, the use of predictive
[47]. These online platforms provide essential binding pocket tools without validation proves to be controversial. Thus, the
information, which could determine the druggability of given reliability of the results is questioned at every stage.
biological targets. Druggability assessment methods utilize a wide range of
algorithms, as discussed. It may be noted that according to the
Ligand-based druggability assessment choice of predictors and parameters, bias results may be ob-
tained. A geometry-based method may find the most suitable
In ligand-based druggability assessment methods, the pocket based solely on the orientation and shape of the pro-
druggability of a therapeutic target is based on a knowledge tein, whereas the energy-based method may find greater mo-
of endogenous compounds that could bind to the target and lecular forces in a different pocket of the same protein. It is,
are yet to reach clinical trials [34]. Several databases available therefore, essential to choose the appropriate predictive soft-
online contain sufficient information about several ligands ware when identifying potentially druggable pockets of a pro-
that could be used in predicting the druggability of a given tein since the principle and level of accuracy of each predictive
target. Notable examples of these online platforms include model vary. For instance, in binding pocket prediction, several
BindingDB [48], PubChem [49], SwissTargetPrediction improved approaches have been developed over the years,
[50], TargetHunter [51], Guide to pharmacology [52], since many of the existing prediction provide varying results.
National Center for Biotechnology Information (NCBI) [53], One such improved binding pocket methods is the
and Protein data bank [54]. Metapocket method [44], which combined four different
methods, namely, SURFNET [60], PASS [61], LIGSITE (cs)
Precedence-based druggability assessment [62], and Q-SiteFinder [63] towards improving its binding
pocket prediction success rate. Therefore, although most of
In employing precedence-based methods for assessing the various algorithms that have been employed by many
druggability, knowledge of established proteins or drug targets existing druggability prediction models have been validated
is required. Precedence-based methods also rely on a knowl- against different training sets, a comparative study of ten rep-
edge of chemical molecules that are already undergoing clin- resentative prediction methods suggest that, of the top three
ical trials [55]. This method demonstrates the highest level of performing methods, 30% of clinically known binding sites
druggability as the given target has already been proven to be were not detected with existing prediction tool [64]. Another
safe. However, precedence-based methods of assessing shortfall of druggable models is that they do not adequately
druggability are still not a guarantee of success for different account for dynamic protein-ligand binding interactions [35].
and or product profiles [42, 56]. Various databases contain Thus, the relevance of binding site target interactions further
information about compounds, targets, and published works indicates that although binding pocket prediction methods are
that could be employed in precedence-based druggability as- useful in identifying putative druggable sites, they have limi-
sessments. Notable examples of some of these freely available tations and are thus not allowed for validation analysis.
databases include DrugCentral [57], Clinicaltrials.gov [58], Therefore, the further integration of druggability information
and Drugbank [59]. As of February 2020, the latest release with other parameters, including the lead compound’s drug-
of DrugBank (2020-01-03) alone contained about 13,491 likeness, would provide a more effective and holistic mecha-
drug entries, including 2641 approved small molecule drugs, nism of disease management.
J Mol Model (2020) 26:120 Page 5 of 11 120

Identification of lead molecules by assessing Physiochemical properties and pharmacokinetic


drug-likeness properties in determining drug-likeness

The concept of drug-likeness has advanced extensively over Assessing the pharmacokinetic and physicochemical proper-
the years, taking into consideration the structural, physico- ties of compounds in determining their drug-likeness occurs in
chemical, biochemical, pharmacokinetic (PK), and toxicity the early stages of the drug discovery process. This process
characteristics of a compound; thus, a drug-like compound takes into consideration structural features and physicochem-
possesses sufficiently acceptable ADME properties, as well ical properties that complement a given pharmacophore with-
as sufficiently acceptable toxicity properties [65, 66]. out affecting it. These pharmacokinetic properties inform the
Understanding these properties has become an integral part behavior of compounds in the body with regard to absorption,
of drug discovery and has enabled the accurate selections of distribution, and excretion. Early pharmacokinetics and phys-
hits that are suitable starting points for the identification of icochemical assessment relied on molecular properties such as
new clinical candidates. These properties also allow a polarity (polar surface area (PSA) and topical surface area),
streamlining of drug discovery efforts towards the identifica- number of aromatic rings, number of heavy atoms, lipophilic-
tion of a compound that possesses a higher chance of pharma- ity, molecular mass, number of chiral centers, number of hy-
cokinetic success and overall safety. drogen bond donors and acceptors, and number of rotatable
The time-consuming nature and enormous resources re- bonds. However, recently these have been expanded to in-
quired in performing in vivo studies to determine the drug- clude complex properties such as metabolic stability, perme-
likeness of a compound have presented computational tech- ability, transporters, solubility, physiologically based pharma-
niques as a viable alternative. Christopher Lipinski and co- cokinetics, and pharmacokinetic/pharmacodynamic modeling
workers reported the earliest report on decisive criteria for [78–80]. A combination of all these complex properties allows
identifying drug-like compounds at Pfizer in the mid-1990s for the optimization and identification of drug-like molecules
[67]. Since then, several reports have attempted to describe the while taking into consideration safety, selectivity, efficacy,
drug-likeness of compounds, including the report by and pharmacokinetics [81]. Based on these molecular proper-
Sadowski et al. [68] in 1998; in which, for the first time, a ties, several guidelines for predicting drug-likeness of chemi-
scoring function was employed in predicting the drug-likeness cal molecules have been proposed, notably, the Lipinski’s rule
of the compounds using neural networks (NN). Subsequently, of five [67], Veber rules [82], Waring drug candidate perme-
many other reports have also employed neural networks in ability [83], and golden triangle rules [84], among numerous
prediction of drug-likeness [69–71]. Other computer-based other existing guidelines for the prediction of drug-likeness.
approaches and molecular descriptors that have also been
employed over the years in the prediction of drug-likeness Lipinski’s rule of five (RO5)
include recursive partitioning (RP) [72, 73], support vector
machine (SVM) [69, 74], and genetic algorithm (GA) [75, As one of the earliest guidelines proposed for predicting the
76], among others. Over the years, these techniques have drug-likeness of compounds, the RO5 emerged from the de-
gained prominence in their ability to predict drug-likeness of position criteria where analysis of drug-likeness is based on
compounds from nondrug compounds satisfactorily. This lipophilicity, molecular weight, and counts of hydrogen bond
eventual incorporation of in silico techniques in predicting acceptors and hydrogen bond donors [85, 86]. According to
drug-likeness of compounds has allowed for the analysis of Lipinski, compounds with sufficiently acceptable ADME/T
large compound libraries to identify compounds that could be properties (thus a molecular weight not greater than 500 Da,
subsequently prioritized for synthesis, in vitro and in vivo, and log P not greater than 5, hydrogen-bond donor not more than 5
for thorough experimental evaluation. In silico approaches and hydrogen bond acceptor (nitrogen and oxygen atoms) not
have thus minimized time and resources that would have been more than 10) to survive through the Phase I clinical trials
wasting in assessing nondrug molecules. Several molecular were considered drug-like. As a guideline, the Lipinski’s
descriptors have been applied in profiling chemical com- RO5 is conceptually simple and straightforward to implement,
pounds towards their drug-likeness. A report by Feher and hence its widespread adoption. However, it does not apply to
Schmidt succinctly highlights several molecular descriptors natural products or substrates of biological transporters [87].
that can be used in characterizing chemical compounds as Based on Lipinski’s assumptions, chemical molecules that
employed in their study [77]. Nonetheless, these are usually violate these rules are less likely to possess many desirable
of little practical help to medicinal chemist since the assess- characteristics of drugs. Nonetheless, there are plenty of ex-
ment of drug-likeness with these molecular descriptors is usu- amples available for RO5 violation among the existing drugs,
ally based on any type of countable feature or computable notably antibiotics, antifungals, vitamins, and cardiac glyco-
property instead of profiling based on biologically relevant sides, but yet these are orally bioavailable [88]. These classes
parameters. of compounds are orally bioavailable because they possess
120 Page 6 of 11 J Mol Model (2020) 26:120

groups that act as substrates for transporters [88]. In RO5, the hydrogen-bond donors, molecular weight, rotatable bonds,
consideration of the number of bonds is due to their direct logD, PSA, and logP. Their report concluded that close to
correlation with an increase in aqueous solubility and a de- property limits, the probability of obtaining high permeability
crease in lipid bilayer permeability. On the other hand, molec- compounds was lower. Their report also showed that the per-
ular weight directly correlates with the size of the compounds, meability of the compounds varies over different ranges of
since large molecular weighted compound could hinder the molecular weight.
diffusion of molecules across lipid bilayer membranes as well
as reduce the molecule solubility. High lipophilicity (logP > 5)
often contributes to high metabolic turnover, aqueous solubil- In silico techniques in predicting
ity, and reduced intestinal absorption. However, high lipophi- drug-likeness
licity of compounds is also associated with an increased risk of
promiscuity and toxicity since these compounds tend to bind The advent of computer-aided drug design has influenced a
to hydrophobic biological targets other than the desired target. paradigm shift towards incorporating computational tech-
niques in identifying drug-like compounds while eliminating
Golden triangle rules compounds that may be unsuccessful in the drug discovery
process. Over the years, experiment filters have been at the
The golden triangle as criteria for selecting drug-like com- forefront in the drug discovery process as the reliable methods
pounds takes into consideration the metabolic stability and are isolating drug-like compounds. However, these conven-
permeability of the compound [84]. In the report by Johnson tional approaches are time demanding and resource-intensive
et al., structural properties that were assessed included molec- [28, 91]. As such, computational approaches have gained
ular weight and lipophilicity (log D7.4) since these parameters prominence in the last two decades as reliable approaches in
affected almost all other drug-likeness properties, which could the prediction of drug-like compounds with much ease.
represent other molecular properties such as rotatable bonds, Computational techniques have typically been involved the
hydrogen bond acceptors and donors, molecular volume, het- filtering out of nondrug compounds as well as compounds
eroatoms, and PSA [84, 89]. Lipophilicity and molecular regarded as unfit for screening purposes by taking into con-
weight are also good predictors of the survival of compounds sideration fundamental physiochemical and pharmacokinetic
in development and are also compatible with chemistry tools properties [29, 92, 93]. In recent decades, several computa-
ligand efficiency (LE) and ligand-lipophilicity efficiency tional software and online platforms have been developed in
(LLE) [90]. predicting essential molecular properties of compounds to-
Drug-like compounds under the golden triangle rule, there- wards the identification of drug molecules with promising
fore, possess high membrane permeability and exhibit favor- potential of achieving success in the drug discovery process,
able metabolic stability if they have a triangle baseline, mo- as highlighted in Table 1.
lecular weight of 200 and lipophilicity of − 2 < logD7.4 < 5, Nonetheless, the challenge with many of such tools is the
and a triangle apex, molecular weight of 450 and logD7.4 = unavailability of validating sets of compounds; thus, some
1.0–2.0. compounds are likely not to fall under any of the criteria being
used by the computational tool. Also, many of these compu-
Waring drug candidate permeability guideline tational tools identify drug-like compounds based on features
of existing drugs; hence, drug-like compounds with new prop-
The membrane permeability of small molecule compounds is erties could easily be mistaken as nondrug compounds [91,
a crucial metric in the drug design pipeline that is essential in 95]. Therefore, these computational tools play an essential
the overall potency and in vivo efficacy of the compounds role in the early stages of the drug discovery process, where
since membrane permeability ensures that the compounds they aid in the prioritization compound for further experimen-
reach their intended target. As such, using membrane perme- tal exploration and other techniques such as high throughput
ability as a parameter, the drug-likeness of compounds could screening.
be predicted. Using permeability as a parameter, Waring et al.
[83] propounded a guideline for the prediction of highly per-
meable drug-like compounds with particular consideration of Conclusions
the apparent permeability (Papp) of the compounds. In their
report, the Papp of the studied compounds was set within a In this review, the fundamental principles and the central con-
range of < 100 nm/s (low) and > 100 nm/s (high). In generat- cepts of druggability and drug-likeness have been briefly com-
ing the guideline, the set range of Papp was statistically municated. There are vast amounts of information and opin-
contrasted with other molecular property predictors such as ions in the assessment and analysis of both the drug-likeness
total hydrogen bonding, hydrogen bond acceptors and of a molecule and the identification of its biological target. In
J Mol Model (2020) 26:120 Page 7 of 11 120

Table 1 In silico tools utilized in the prediction of molecule drug-likeness [94]

In silico tool Function

QikProp (https://www.schrodinger.com/qikprop) distributed by Schrodinger Predicts and evaluates the ADME properties of
thousands of compounds within a short time
VolSurf+ (https://www.moldiscovery.com/software/volsurf/) distributed by Molecular Discovery Ltd. A computational tool that calculates and
(moldiscovery.com) predicts essential ADME properties of
compounds using predictive 3D molecular
interaction energy grid maps
MedChem Studio (https://www.simulations-plus.com/software/admetpredictor/medchem-studio/) A platform commonly used by medicinal
distributed by Simulation Plus, Inc. chemists and computational chemists for
identifying lead compounds. It could also be
employed for the optimization of the
identified lead compounds as well as the
prediction of ADME properties. MedChem
studio is a complete computer-aided drug
design package that can also be used for
ligand-based drug design and the
categorization of libraries of compounds
ADMET Predictor (https://www.simulations-plus.com/software/admetpredictor/) distributed by A computational software that allows for the
Simulations Plus, Inc prediction of essential physiochemical
ADMET properties as well crucial CYP
metabolism kinetics of molecular structures.
ADMET Predictor can also be used to
generate predictive models of entirely new
properties from the user’s library of structures
using the inbuilt ADMET Modeler module
ADMEWORKS ModelBuilder (https://www.fqs.pl/en/products) distributed by Fujitsu ADMEWORKS ModelBuilder generates
QSAR/QSPR models that can be employed
in predicting physicochemical and
pharmacokinetic properties of compounds
ADMEWORKS Predictor (https://www.fqs.pl/en/products) Distributed by Fujitsu An in silico tool for evaluating and predicting
essential pharmacological properties of
compounds by employing QSAR-based
virtual screening
IMPACT-F (http://www.pharmainformatic.com/html/impact-f.html) developed by PharmaInformatic, A reliable tool made up of QSAR models for
Germany evaluating the human oral bioavailability of
drug candidates
ADMET Modeler (http://www.simulations-plus.com/software/admetpredictor/admet-modeler/) A plugin of the ADMET Predictor that
distributed by Simulation Plus, Inc. addresses the tedious process of building
reliable QSAR/QSPR models from
experimentally derived data sets by
automating the entire process
ACD/PhysChem Suite (https://www.acdlabs.com/products/percepta/) distributed by ACD/Labs Employs fragment-based models to rapidly
predict various physicochemical and
pharmacokinetic properties of compounds
such as lipophilicity (logD, logP), aqueous
solubility, and pKa values, among other
molecular descriptors
ACD/ADME Suite (https://www.acdlabs.com/products/percepta/index.php) distributed by ACD/Labs A computational tool that predicts the ADME
properties of compounds by estimating
various molecular descriptors such as oral
bioavailability, blood-brain barrier
permeability, and absorption, among others
ACD/DMSO (https://www.acdlabs.com/products/percepta/index.php) distributed by ACD/Labs Predicts solubility in DMSO solution
Discovery Studio ADMET Software (https://www.3dsbiovia. Predicts the ADMET properties for compounds
com/products/collaborative-science/biovia-discovery-studio/qsar-admet-and-predictive-toxicology.
html) distributed by Accelrys
MedChem Designer (http://www.simulations-plus.com/software/medchem-designer/) distributed by ADMET properties predicting tools that
Simulation Plus, Inc. incorporate chemical structure drawing
features
120 Page 8 of 11 J Mol Model (2020) 26:120

Table 1 (continued)

In silico tool Function

Filter-it (http://silicos-it.be.s3-website-eu-west-1.amazonaws.com/software/filter-it/1.0.2/filter-it. Based on pre-programmed molecular


html) distributed by Silicos-it descriptors, the Filter-it tool screens out
compounds with undesirable properties from
a set of compounds
PreADME (https://preadmet.bmdrc.kr/) hosted by Yonsei University, Seoul, Republic of Korea An in silico tool for predicting the drug-likeness
of compounds as well as the calculation of
molecular descriptors
PrologP/PprologD (https://www.compudrug.com/?q=node/42) distributed by CompuDrug Employs both neural network methods and
linear methods towards the prediction of the
lipophilicity (logP/logD) of compounds
Web Services for predicting drug-likeness
Chemicalize (https://chemicalize.com/) provided by ChemAxon Online service for the prediction of
physicochemical and pharmacokinetic
properties such as logP, molecular weight,
hydrogen bond donors, and acceptors. It also
calculates molecular properties such as
tautomers and PSA
SwissADME (http://www.swissadme.ch/) developed and maintained by the Molecular Modeling A robust and reliable online service for
Group of the SIB Swiss Institute of Bioinformatics predicting drug-likeness of compounds as
well as their pharmacokinetic and
physicochemical properties based on reliable
in-house methods notably the bioavailability
rader, BOILED-Egg, synthetic accessibility
score, and iLOGP
AqualSol (http://cdb.ics.uci.edu/cgibin/tools/AquaSolWeb.py) provided by the University of Employs UG-RNN ensembles to predict
California, Irvine aqueous solubility of compounds
Molinfo (http://cdb.ics.uci.edu/cgibin/tools/MolInfoWeb.py) provided by the University of An online platform for predicting various
California, Irvine essential molecular properties of compounds
that are of interest to medicinal chemists
DrugMint (http://crdd.osdd.net/oscadd/drugmint/) maintained by Indraprastha Institute of Web server predicting the drug-likeness of
Information Technology compounds
DrugLogit (http://hermes.chem.ut.ee/~alfx/druglogit.html) maintained by the Institute of Chemistry, An online that predicts the chances of chemical
University of Tartu, Estonia structural being categorized as a drug or a
nondrug compound web service to predict the
probability of a compound. DrugLogit also
performs the classification of compounds
based on the disease category
AdmetSAR (http://lmmd.ecust.edu.cn/admetsar1/) maintained by Shanghai Key Laboratory of New A web server that predicts ADMET properties
Drug Design, School of Pharmacy, East China University of Science and Technology of compounds. A peculiar feature of this
server is its ability to predict about 50
ADMET endpoints using a
cheminformatics-based tool called
ADMET-Simulator
PkCSM (http://biosig.unimelb.edu.au/pkcsm/) developed by the University of Cambridge, UK This online service employs graph-based
signatures to calculate the pharmacokinetic
properties of compounds
MODEL-Molecular Descriptor Lab Based on 3D structures, this online service
(http://jing.cz3.nus.edu.sg/cgi%E2%80%90bin/model/model.cgi) maintained by the University of calculates the physiochemical and structural
Singapore properties of compounds
OSIRIS Property Explorer (https://www.organic-chemistry.org/prog/peo/) maintained by the Virtual This web service allows for the calculation of
Computational Chemistry Laboratory the drug-like properties of drawn structures as
well as the prediction of toxicity profiles. It
forms part of the Actelion’s in-house
substance registration system
Property calculator (https://mcule.com/apps/property-calculator/) provided by mcule Generate the physicochemical properties of
chemical structures
PreADMET (https://preadmet.bmdrc.kr/) maintained by the Bioinformatics and molecular design An online web service that calculates the
research center of Yonsei University, Korea ADME properties of compounds. It also
J Mol Model (2020) 26:120 Page 9 of 11 120

Table 1 (continued)

In silico tool Function

generates a drug-like compound library using


computational methods
Databases
ADME-AP (http://bidd.nus.edu.sg/group/admeap/admeap.asp) maintained by the Dept. ADME-AP is a database that houses a vast
Computational Science. NUS library of data about drug targets such as
tissue distributions, functions, substrates, and
known ADME linked targets. It simplifies the
search for drug ADME related proteins
The ADME database (https://www.fujitsu. An extensive database of structurally different
com/jp/group/kyushu/en/solutions/industry/lifescience/admedatabase/) distributed by Aureus compounds coupled with known ADME
features

this study, however, computational approaches were catego- 6. Taboureau O, Nielsen S, Audouze K (2011) ChemProt: a disease
chemical biology database. Nucleic Acids Res 39:D367–D372
rized and briefly explained. This provided a deeper apprecia-
7. Dixon S, Stockwell B (2009) Identifying druggable disease-
tion of how online predictive tools and techniques streamline modifying gene products. Curr Opin Chem Biol 13:549–555
the drug discovery process by not wasting a copious number 8. Scannell J, Blanckley A, Boldon H et al (2012) Diagnosing the
of years with “trial and error” experimental investigations. decline in pharmaceutical R&D efficiency. Nat Rev Drug Discov
Although these “click of a button” procedures may shorten 11:191–200
9. Hingorani A, Kuan V, Finan C et al (2019) Improving the odds of
the rational drug discovery process, it is always important to
drug development success through human genomics: modelling
keep in mind that predictive data is only ever “potential” in- study. Sci Rep 9:18911
formation, until validated with experimental findings. We de- 10. Hay M, Thomas D, Craighead J, Economides C, Rosenthal J (2014)
duce that the synergistic capabilities of predictive and exper- Clinical development success rates for investigational drugs. Nat
imental investigations may lead to a greater understanding of Biotechnol 32:40–51
11. Schmidtke P, Barril X (2010) Understanding and predicting
drug resistance in diseases, thereby allowing for more rapid druggability. A high-throughput method for detection of drug bind-
and efficient treatment regimes. ing sites. J Med Chem 53(15):5858–5867
12. Vistoli G, Pedretti A, Testa B (2008) Assessing drug-likeness - what
Acknowledgments The authors appreciate the College of Health are we missing? Drug Discov Today 13(7–8):285–294
Sciences, University of KwaZulu-Natal for financial and infrastructural 13. Kozakov D, Hall DR, Napoleon RL, Yueh C, Whitty A, Vajda S
support, while we also thank the Center for High-Performance (2015) New frontiers in druggability. J Med Chem 58(23):9063–
Computing (CHPC, www.chpc.ac.za) Cape-Town, South Africa, for 9088
providing computational resources. 14. Oprea TI, Bologa CG, Brunak S, Campbell A, Gan GN, Gaulton A
et al (2018) Unexplored therapeutic opportunities in the human
Compliance with ethical standards genome. Nat Rev Drug Discov 7:317–332
15. Finan C, Gaulton A, Kruger FA, Lumbers RT, Shah T, Engmann J
et al (2017) The druggable genome and support for target identifi-
Conflict of interest The authors declare that they have no conflict of
cation and validation in drug development. Sci Transl Med 9(383):
interest.
eaag1166
16. Hopkins AL, Groom CR (2002) The druggable genome. Nat Rev
Drug Discov 1(9):727–730
17. Cheng AC, Coleman RG, Smyth KT, Cao Q, Soulard P, Caffrey DR
References et al (2007) Structure-based maximal affinity model predicts small-
molecule druggability. Nat Biotechnol 25(1):71–75
1. Dang CV, Reddy EP, Shokat KM, Soucek L (2017) Drugging the 18. Egner U, Hillig RC (2008) A structural biology view of target
“undruggable” cancer targets. Nat Rev Cancer 17(8):502–508 druggability. Expert Opin Drug Discovery 3(4):391–401
2. Galdeano C (2017) Drugging the undruggable: targeting challeng- 19. Sheridan RP, Maiorov VN, Holloway MK, Cornell WD, Gao YD
ing E3 ligases for personalized medicine. Future Med Chem 9(4): (2010) Drug-like density: a method of quantifying the “bindability”
347–350 of a protein target based on a very large set of pockets and drug-like
3. Zhang ZY (2017) Drugging the undruggable: therapeutic potential ligands from the protein data bank. J Chem Inf Model 50(11):2029–
of targeting protein tyrosine phosphatases. Acc Chem Res 50(1): 2040
122–129 20. Rathi PC, Ludlow RF, Hall RJ, Murray CW, Mortenson PN,
4. Machado D, Girardini M, Viveiros M, Pieroni M (2018) Verdonk ML (2017) Predicting “hot” and “warm” spots for frag-
Challenging the drug-likeness dogma for new drug discovery in ment binding. J Med Chem 60(9):4036–4046
tuberculosis. Front Microbiol 9:1367 21. Hajduk PJ, Huth JR, Fesik SW (2005) Druggability indices for
5. Sakharkar M, Sakharkar K (2007) Targetability of human disease protein targets derived from NMr-based screening data. J Med
genes. Curr Drug Discov Technol 4:48–58 Chem 48(7):2518–2525
120 Page 10 of 11 J Mol Model (2020) 26:120

22. Seco J, Luque FJ, Barril X (2009) Binding site detection and Available from: http://www.liebertonline.com/doi/abs/10.1089/
druggability index from first principles. J Med Chem 52(8):2363– omi.2009.0045
2371 45. Hussein HA, Borrel A, Geneix C, Petitjean M, Regad L, Camproux
23. Halgren TA (2009) Identifying and characterizing binding sites and AC (2015) PockDrug-server: a new web server for predicting pock-
assessing druggability. J Chem Inf Model 49(2):377–389 et druggability on holo and apo proteins. Nucleic Acids Res
24. Hajduk PJ, Huth JR, Tse C (2005) Predicting protein druggability. 43(W1):W436–W442
Drug Discov Today 10(23–24):1675–1682 46. Hernandez M, Ghersi D, Sanchez R (2009) SITEHOUND-web: a
25. Krasowski A, Muthas D, Sarkar A, Schmitt S, Brenk R (2011) server for ligand binding site identification in protein structures.
DrugPred: a structure-based approach to predict protein Nucleic Acids Res 37:413–416
druggability developed using an extensive nonredundant data set. 47. Koscielny G, An P, Carvalho-Silva D, Cham JA, Fumis L,
J Chem Inf Model 51(11):2829–2842 Gasparyan R et al (2017) Open targets: a platform for therapeutic
26. Volkamer A, Kuhn D, Rippmann F, Rarey M (2012) Dogsitescorer: target identification and validation. Nucleic Acids Res 45(D1):
a web server for automatic binding site prediction, analysis and D985–D994
druggability assessment. Bioinformatics. 28(15):2074–2075 48. Liu T, Lin Y, Wen X, Jorissen RN, Gilson MK (2007) BindingDB: a
27. Perola E, Herman L, Weiss J (2012) Development of a rule-based web-accessible database of experimentally determined protein-
method for the assessment of protein druggability. J Chem Inf ligand binding affinities. Nucleic Acids Res 35(D4):198–201
Model 2(4):1027–1038 49. Kim S, Chen J, Cheng T, Gindulyte A, He J, He S et al (2019)
28. Walters W, Ajay A, Murcko M (1999) Recognizing molecules with PubChem 2019 update: improved access to chemical data.
drug-like properties. Curr Opin Chem Biol 3(4):384–387 Nucleic Acids Res 47(D1):D1102–D1109
29. Walters W, Stahl M, Murcko M (1998) Virtual screening – an over- 50. Gfeller D, Grosdidier A, Wirth M, Daina A, Michielin O, Zoete V
view. Drug Discov Today 3:160–178 (2014) SwissTargetPrediction: a web server for target prediction of
30. Bickerton GR, Paolini GV, Besnard J, Muresan S, Hopkins AL bioactive small molecules. Nucleic Acids Res 42(W1):32–38
(2012) Quantifying the chemical beauty of drugs. Nat Chem 4: 51. Wang L, Ma C, Wipf P, Liu H, Su W, Xie X-Q (2013) TargetHunter:
90–98 An in silico target identification tool for predicting therapeutic po-
31. Oprea TI (2000) Property distribution of drug-related chemical da- tential of small organic molecules based on Chemogenomic data-
tabases. J Comput Aided Mol Des 14(3):251–264 base. AAPS J 15(2):395–406
32. Leeson PD, Springthorpe B (2007) The influence of drug-like con- 52. Harding SD, Sharman JL, Faccenda E, Southan C, Pawson AJ,
cepts on decision-making in medicinal chemistry. Nat Rev Drug Ireland S et al (2017) The IUPHAR/BPS guide to pharmacology
Discov 6(11):881–890 In 2018: updates and expansion to encompass the new guide to
33. Katsila T, Spyroulias GA, Patrinos GP, Matsoukas MT (2016) immunopharmacology. Nucleic Acids Res 46(D1):D1091–D1106
Computational approaches in target identification and drug discov- 53. Bethesda (MD) (1988) National Library of Medicine (US).
ery. Comput Struct Biotechnol J 14(1):177–184
National Center for Biotechnology Information (NCBI). Accessed
34. Barril X (2013) Druggability predictions: methods, limitations, and
[11/03/2018]
applications. Wiley Interdiscip Rev Comput Mol Sci 3(4):327–338
54. Berman HM, Westbrook J, Feng Z, Gilliland G, Bhat TN, Weissig
35. Cheng T, Hao M, Takeda T, Bryant SH, Wang Y (2017) Large-scale
H et al (2000) The Protein Data Bank 28(1):235–42
prediction of drug-target interaction: a data-centric review. AAPS J
55. Loving KA, Lin A, Cheng AC (2014) Structure-based druggability
19(5):1264–1275
assessment of the mammalian structural proteome with inclusion of
36. Neal KB, Mahmoud ES (2017) Can we rely on computational pre-
light protein flexibility. PLoS Comput Biol 10(7):321–329
dictions to correctly identify ligand binding sites on novel protein
56. Edfeldt F, Folmer R, Breeze A (2011) Fragment screening to predict
drug targets? Assessment of binding site prediction methods and a
druggability (ligandability) and lead discovery success. Drug
protocol for validation of predicted binding sites. Cell Biochem
Discov Today 16(7–8):284–287
Biophys 75(1):15–23
37. Sun T, Lai L, Pei J (2018) Analysis of protein features and machine 57. Ursu O, Holmes J, Bologa CG, Yang JJ, Mathias SL, Stathias Vet al
learning algorithms for prediction of druggable proteins. Quant Biol (2019) DrugCentral 2018: An update. Nucleic Acids Res 47(D1):
6(4):334–343 D963–D970
38. Li YH, Yu CY, Li XX, Zhang P, Tang J, Yang Q et al (2018) 58. Tse T, Fain KM, Zarin DA (2018) How to avoid common problems
Therapeutic target database update 2018: enriched resource for fa- when using CliicalTrials.gov in research. Bmj. 361(4):1452–1459
cilitating bench-to-clinic research of targeted therapeutics. Nucleic 59. Wishart DS, Feunang YD, Guo AC, Lo EJ, Marcu A, Grant JR et al
Acids Res 46(D1):D1121–D1127 (2018) DrugBank 5.0: a major update to the DrugBank database for
39. Jiang Q, Wang J, Wu X, Ma R, Zhang T, Jin S et al (2015) 2018. Nucleic Acids Res 46(D1):D1074–D1082
LncRNA2Target: a database for differentially expressed genes after 60. Laskowski R (1995) SURFNET: a program for visualizing molec-
IncRNA knockdown or overexpression. Nucleic Acids Res ular surfaces, cavities, and intermolecular interactions. J Mol Graph
43(D1):D193–D196 13:323–330
40. Griffiths-Jones S (2010) MiRBase: MicroRNA sequences and an- 61. Brady G, Stouten P (2000) Fast prediction and visualization of
notation. Curr Protoc Bioinformatics 34(29):1291–12910 protein binding pockets with PASS. J Comput Aided Mol Des 14:
41. Kandoi G, Acencio ML, Lemke N (2015) Prediction of druggable 383–401
proteins using machine learning and systems biology: a mini-re- 62. Huang B, Schroeder M (2006) LIGSITEcsc: predicting ligand bind-
view. Front Physiol 6(9):54–65 ing sites using the connolly surface and degree of conservation.
42. Wyatt PG, Gilbert IH, Read KD, Fairlamb AH (2011) Target vali- BMC Struct Biol 6:19
dation: linking target and chemical properties to desired product 63. Laurie A, Jackson R (2005) Q-SiteFinder: an energy-based method
profile. Curr Top Med Chem 11(10):1275–1283 for the prediction of protein-ligand binding sites. Bioinformatics.
43. Fauman E, Rai B, Huang E (2011) Structure-based druggability 21(9):1908–1916
assessment—identifying suitable targets for small molecule thera- 64. Leis S, Schneider S, Zacharias M (2010) In Silico prediction of
peutics. Curr Opin Chem Biol 15(4):463–468 binding sites on proteins. Curr Med Chem 17(15):1550–1562
44. Huang B (2009) MetaPocket: a meta approach to improve protein 65. Kerns E, Di L (2008) Drug-like properties: concepts, structure de-
ligand binding site prediction. OMICS 13(4):325–330 [Internet]. sign and methods, vol 552. Academic Press, Burlington
J Mol Model (2020) 26:120 Page 11 of 11 120

66. Schneider G Prediction of Drug-Like Properties. Madame Curie 81. Li D, Kerns E, Carter G (2009) Drug-like property concepts in
Bioscience Database. Landes Bioscience, Austin pharmaceutical design. Curr Pharm Des 15:2184–2194
67. Lipinski C, Dominy B, Feeney P (1997) Experimental and compu- 82. Veber D, Johnson S, Cheng H, Smith B, Ward K, Kopple K (2002)
tational approaches to estimate solubility and permeability in drug Molecular properties that influence the oral bioavailability of drug
discovery and development settings. Adv Drug Deliv Rev 23:3–25 candidates. J Med Chem 45:2615–2623
68. Sadowski J, Kubinyi H (1998) A scoring scheme for discriminating 83. Waring M (2009) Defining optimum lipophilicity and molecular
between drugs and nondrugs. J Med Chem 41:3325–3329 weight ranges for drug candidates—molecular weight dependent
69. Byvatov E, Fechner U, Sadowski J, Schneider G (2003) lower logD limits based on permeability. Bioorg Med Chem Lett
Comparison of support vector machine and artificial neural net- 19:2844–2851
work systems for drug/nondrug classification. J Chem Inf 84. Johnson T, Dress K, Edwards M (2009) Using the golden triangle to
Comput Sci 43:1882–1889 optimize clearance and oral absorption. Bioorg Med Chem Lett 19:
70. Takaoka Y, Endo Y, Yamanobe S, Kakinuma H, Okubo T, 5560–5564
Shimazaki Y et al (2003) Development of a method for evaluating 85. Lipinski C (2000) Drug-like properties and the causes of poor sol-
drug-likeness and ease of synthesis using a data set in which com- ubility and poor permeability. J Pharmacol Toxicol Methods 44(1):
pounds are assigned scores based on chemists’ intuition. J Chem 235–249
Inf Comput Sci 43:1269–1275
86. Lipinski CA (2016) Rule of five in 2015 and beyond: target and
71. Ajay A, Walters W, Murcko M (1998) Can we learn to distinguish
ligand structural limitations, ligand chemistry structure and drug
between “drug-like” and “nondrug-like” molecules? J Med Chem
discovery project decisions. Adv Drug Deliv Rev 101:34–41
41:3314–3324
72. Wagener M, van Geerestein V (2000) Potential drugs and nondrugs: 87. Lipinski CA, Lombardo F, Dominy BW, Feeney PJ (2012)
prediction and identification of important structural features. J Experimental and computational approaches to estimate solubility
Chem Inf Comput Sci 40:280–292 and permeability in drug discovery and development setting. Adv
73. Schneider N, Jaeckels C, Andres C, Hutter M (2008) Gradual in Drug Deliv Rev 64:4–17
silico filtering for druglike substances. J Chem Inf Model 48:613– 88. Kadam R, Roy N (2007) Recent trends in drug-likeness prediction:
628 a comprehensive review of in silico methods. Indian J Pharm Sci
74. Zernov V, Balakin K, Ivaschenko A, Savchuk N, Pletnev I (2003) 69(5):609–615
Drug discovery using support vector machines. The case studies of 89. Bhal S, Kassam K, Peirson I, Pearl G (2007) The rule of five
drug-likeness, agrochemical-likeness, and enzyme inhibition pre- revisited: applying log D in place of log P in drug-likeness filters.
dictions. J Chem Inf Model 43:2048–2056 Mol Pharm 4:556–560
75. Gillet V, Khatib W, Willett P, Fleming P, Green D (2002) 90. Murphy RB, Philipp DM, Friesner RA (2000) A mixed quantum
Combinatorial library design using a multiobjective genetic algo- mechanics/molecular mechanics (QM/MM) method for large-scale
rithm. J Chem Inf Comput Sci 42:375–385 modeling of chemistry in protein environments. J Comput Chem
76. Gillet V, Willett P, Bradshaw J (1998) Identification of biological 21(16):1442–1457
activity profiles using substructural analysis and genetic algorithms. 91. Clark D, Pickett S (2000) Computational methods for the prediction
J Chem Inf Comput Sci 38:165–179 of “drug-likeness.”. Drug Discov Today 5(2):49–58
77. Feher M, Schmidt J (2003) Property distribution: differences be- 92. Lewis R, Mason J, McLay I (1997) Similarity measures for rational
tween drugs, natural products, and molecules from combinatorial set selection and analysis of combinatorial libraries: the diverse
chemistry. J Chem Inf Comput Sci 43:218–227 property-derived (DPD) approach. J Chem Inf Comput Sci 37:
78. Rowland M, Peck C, Tucker G (2011) Physiologically-based phar- 599–614
macokinetics in drug development and regulatory science. Annu 93. Rishton G (1997) Reactive compounds and in vitro false positives
Rev Pharmacol Toxicol 51:45–73 in HTS. Drug Discov Today 2:382–384
79. Gabrielsson J, Green A (2009) Quantitative pharmacology or phar- 94. Bioinformatics SI of Click2Drug (2013) p 1–10
macokinetic pharmacodynamic integration should be a vital com- 95. Polinsky A (1999) Combichem and chemoinformatics. Curr Opin
ponent in integrative pharmacology. J Pharmacol Exp Ther 331: Drug Discov Devel 2:197–203
767–774
80. Wager T, Hou X, Villalobos A (2010) Moving beyond rules: the
development of a central nervous system multiparameter optimiza- Publisher’s note Springer Nature remains neutral with regard to jurisdic-
tion (CNS MPO) approach to enable alignment of druglike proper- tional claims in published maps and institutional affiliations.
ties. ACS Chem Neurosci 1:435–449

You might also like