Computer-aided drug design (CADD) often involves virtual screening (VS) of large compound datasets and the availability of such is vital for drug discovery protocols. We assess the bioactivity and “drug-likeness” of a relatively small but structurally diverse dataset (containing >1,000 compounds) from African medicinal plants, which have been tested and proven a wide range of biological activities. The geographical regions of collection of the medicinal plants cover the entire continent of Africa, based on data from literature sources and information from traditional healers. For each isolated compound, the three dimensional (3D) structure has been used to calculate physico-chemical properties used in the prediction of oral bioavailability on the basis of Lipinski’s “Rule of Five”. A comparative analysis has been carried out with the “drug-like”, “lead-like”, and “fragment-like” subsets, as well as with the Dictionary of Natural Products. A diversity analysis has been carried out in comparison with the ChemBridge diverse database. Furthermore, descriptors related to absorption, distribution, metabolism, excretion and toxicity (ADMET) have been used to predict the pharmacokinetic profile of the compounds within the dataset. Our results prove that drug discovery, beginning with natural products from the African flora, could be highly promising. The 3D structures are available and could be useful for virtual screening and natural product lead generation programs.
Citation: Ntie-Kang F, Zofou D, Babiaka SB, Meudom R, Scharfe M, Lifongo LL, et al. (2013) AfroDb: A Select Highly Potent and Diverse Natural Product Library from African Medicinal Plants. PLoS ONE 8(10): e78085. https://doi.org/10.1371/journal.pone.0078085
Editor: Joseph J. Barchi, National Cancer Institute at Frederick, United States of America
Received: June 13, 2013; Accepted: September 6, 2013; Published: October 30, 2013
Copyright: © 2013 Ntie-Kang et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Funding: Financial support is acknowledged from the Chemical and Bioactivity Information Centre, University of Buea, through a Lhasa grant and also from the German Academic Exchange Service to FNK for his stay in Halle, Germany for part of his PhD, as well as from the ICTP through the OEA-AC-071 program. DZ acknowledges an International Foundation for Science, Stockholm, Sweden/Organisation for the Prohibition of Chemical Weapons, The Hague, The Netherlands (IFS/OPCW) grant (N°F/5122-1). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The academic license for this work was generously offered by Schrodinger Inc. This does not alter the authors’ adherence to all the PLOS ONE policies on sharing data and materials.
Drug design and discovery efforts have often resorted to natural sources for hit/lead compound identification –. This is because nature is an enormous source for structurally diverse chemical scaffolds from which drugs can be isolated and/or synthesized –. Moreover, natural products (NPs) are unique in that they are often rich in stereogenic centres and cover segments of chemical space which are typically not occupied by a majority of synthetic molecules and drugs –. In addition, it can be verified that the African flora has a huge potential and remains an interesting reservoir for new drugs targeting a variety of diseases .
Modern drug discovery efforts also incorporate computer-aided approaches like ligand docking, pharmacophore searching, neural networking and binding free energy calculations of potential drugs towards a target receptor. The rationale behind such in silico methods has been to simulate the interaction between a potential drug molecule and its receptor or binding site (often the drug target) using 3D computer models –. Hence, the use of computer modeling in drug discovery otherwise known as computer-aided drug design (CADD) requires a compound library containing 3D structures of potential leads, which need to be screened in silico, with the view of identifying hit compounds. If this effort is successful, then the identified hits could be confirmed as active compounds using screening assays. Such a procedure considerably cuts down the cost of drug discovery and development .
At the moment, efforts towards drug discovery from the African flora have been limited to random screening of extracts and/or bioassay guided fractionation of extracts from medicinal plant materials, based on information obtained from the ethnobotanical uses of the plants . However, these efforts remain below expectations and their applicability as well as their impact, are not felt at the level of local populations. Moreover, most screening efforts are limited to crude extracts and/or in vitro screening, with only little work on clinical development of the identified active molecules. The inclusion of in silico methods of drug discovery into the scene would likely foster efforts towards lead optimization and facilitates the entry of most interesting compounds into clinical trials. However, this process requires the development of databases of 3D structures of compounds, which have been isolated from medicinal plants in Africa. It should be mentioned that some of the chemical structures are available in published articles in internationally recognized peer-reviewed journals. However, locally published data (in MSc and PhD theses as well as in local or national journals) may not be readily available to the wider scientific community. Moreover, the absence of 3D structures heavily hampers in silico virtual screening. In our continuous efforts to build virtual natural product libraries for compounds isolated from the African flora –, it has become necessary to identify selected highly potent compounds from the entire continent, generate their 3D structures and make them available for use in virtual screening campaigns.
With the exponential increase in computer power , it has become possible today to carry out successful virtual screening on huge databases like the ZINC library (currently 19,607,982 purchasable compounds) – in a matter of weeks , . However, such computer power is currently absent in most research laboratories on the African continent, thoughtless of guaranteeing continuous electrical power supplies for long periods of time. This has therefore necessitated the development of relatively small compound library (∼1000 compounds), known to contain NPs from across the continent, which have recorded activities against a wide range of tropical diseases as well as diseases dominant in rich countries (like cancer and hypertension). We therefore present AfroDb, which is available in several file formats, and which could be highly useful in CADD efforts. An evaluation of the potential oral bioavailability has been carried out by Lipinski criteria  in comparison with Dictionary of Natural Products (DNP) . Several parameters related to drug metabolism and pharmacokinetics (DMPK) are also computed. A diversity analysis has also been carried out in comparison with the ChemBridge Diverset library .
Results and Discussion
Origin and Description of Secondary Metabolites
The distribution of the compound collection by geographical region of plant origin is shown in Figure 1. This shows that a majority of the compounds with remarkable biological activities were derived from the Central Africa region (35%), followed by Southern Africa (23%) and East Africa (21%). The known biological activities include very specific descriptions (like inhibition or modulation of known drug targets, e.g., prolyl endopeptidase I inhibition, 11β-hydroxysteroid dehydrogenase inhibition, α-glucosidase inhibition, enhancement of cAMP-regulated chloride conductance of cells expressing CFTRΔF508, and snake venom phosphodiesterase I inhibition), while more unspecific classifications include anti-HIV, antisalmonellal, antimalarial, antileishmanial, antitubercular, antitrypanosomal, antitumour, vasodilator, vasorelaxant and hypertensive effects, estrogenic activities, and activity against Onchocerca gutturosa.
Discussion of Lipinski’s Oral Availability Criteria and Property Distribution
In modern drug discovery, the identification of lead compounds often involves the development of compound libraries with a high level of molecular diversity within the limits of significant “drug-like” properties , . Thus, Lipinski’s criteria , generally referred to as the “rule of Five” (ro5), have been used in the evaluation of the likely oral availability of the compounds within the AfroDb database. In summary, Lipinski’s ro5 defines a likely orally available molecule as one for which the molar weight (MW) ≤500 Daltons (Da), the logarithm of the octanol/water partition coefficient representing the lipophilicity factor (log P) ≤5, the number of hydrogen bond acceptors (HBA) ≤10 and the number of hydrogen bond donors (HBD) ≤5. A fifth rule dealing with the number of rotatable bonds (NRB) is often added to these, such that NRB ≤5. The distributions of the compound MW, log P, HBA and HBD were calculated and used to assess the likely oral availability of AfroDb, as shown in Figure 2. It is noteworthy that natural products exhibit a wide range of flexibility, from rigid conformationally constrained molecules to very flexible compounds. Thus, the number of rotatable bonds (NRB) within the AfroDb library was used as an additional criterion to test for the favourable drug metabolism and pharmacokinetics (DMPK) outcomes. It was observed that 57.8% of the compounds within AfroDb showed no Lipinski violations and 84.3% showed <2 violations (Figure 2A), while the peak of the distribution of the MW lies between 301 and 400 (Figure 2B). The log P distribution showed a Gaussian shaped curve with a peak centred at 3.5 log P units (Figure 2C), while those of the HBA and HBD rose rapidly to peak values at 5 acceptors and 1 donor respectively. Moreover, both HBA and HBD graphs fell rapidly to maximum values of 27 acceptors and 14 donors (Figures 2D and E). The peak value for NRB was between 3 and 4 (Figure 2F), the graph also falling rapidly to a maximum value of 61 rotatable bonds (RBs). This last parameter indicates the high degree of flexibility of some of the NPs within the database. The MW distribution resembled those previously reported for other “drug-like” NP libraries in the literature , , , with only about 9% of MW >500 Da. In addition, only about 16% of the compounds had log P values >5, only about 7% having HBA >10 and only ∼9% had HBD >5. It was observed, however, that some of the compounds had log P values as high as >21 units. The exceptionally large calculated logP values (>14) observed in compounds 1 to 3 could be explained by complex partitioning in long chain aliphatic compounds, which could not be properly taken into account by the algorithm employed in the log P prediction. The three long chain compounds with log P>14 have been shown in Figure 3, with their plant sources and biological activities given in Table 1. The above arguments could be verified by the fact the scatter plots displaying the mutual relationship between the MW versus the other calculated parameters (log P, HBA, HBD and NRB), Figure 4, show the highest densities of points within the Lipinski compliance regions (MW <500, −2< log P<5, HBA <10, HBD <3), and for which NRB <5.
(A, B) Histogram of Lipinski violations as a percentage of the AfroDb data set and molar weight distribution, respectively. (C, D, E, F) Distribution curves of the log P, HBA, HBD and NRB, respectively for the 1,008 compounds currently in AfroDb. For subfigure B, the x-axis label is the lower limit of binned data, e.g. 0 is equivalent to 0 to 100.
(A) The distribution of the calculated log P versus MW, (B) HBA against MW, (C) HBD against MW and (D) NRB versus MW. LCR represents the Lipinski compliant regions.
Comparison with the Dictionary of Natural Products (DNP)
The distributions of the individual parameters for AfroDb and the DNP have been compared (Figure 5). These histograms show only data falling within the Lipinski compliance zones (MW <500, −2≤ log P<5, HBA <10, and HBD <5), the values being expressed as percentage counts of the respective databases. Our analyses showed an enhancement of the distributions of AfroDb over the DNP for Lipinski properties. As for the MW distribution histograms (Figure 5A), both curves show peaks at 301–400 Da, the AfroDb having >19% enhancement in MW for the region 301 to 500 Da. Below this range, the percentages were reduced for the AfroDb when compared to the DNP.
(A) MW, (B) log P, (C) HBA and (D) HBD. DNP in red and AfroDb in blue. For subfigure B, the x-axis label is the lower limit of binned data, e.g. −2 is equivalent to −2 to −1.
It is noteworthy that an enhancement of the MW profile is a desirable factor for a more “drug-like library”, according to Lipinski’s criteria. The proportions of the two databases that satisfy Lipinski’s MW property (<500 Da) were about 73% for the DNP, as compared to about 84% for AfroDb. This showed an enhancement of 11.2% for MW values between 301 and 500 Da of AfroDb over the DNP. The maximum values for the log P distributions were respectively at 3.5 and 2.5 log P units for AfroDb and the DNP (Figure 5B). In addition, a corresponding enhancement of 9.7% for log P values between 2 and 5 units was observed for AfroDb over the DNP. As per the HBA and HBD respectively (Figures 5C–D), AfroDb showed improvements of 17.1% for 3< HBA <8 and 10.8% for 0< HBD <4 over the DNP. The peak of the distribution for the HBA for the AfroDb is at 5 acceptors (19.3%) with a significant increase in 7 or 8 acceptors when compared to the DNP (Figure 5C). Similarly, the peak of the distribution for the HBD for the AfroDb is at 1 acceptor (25.4%) with a significant increase in 3 or 4 donors as compared to the DNP (Figure 5D). The overall summary of the four Lipinski parameters for the two datasets thus reveals that the AfroDb library is more “drug-like” than the DNP, indicating that the chances of finding “lead-like” molecules with improved DMPK properties within a library such as AfroDb are quite significant.
Overall DMPK Compliance of the AfroDb Library
There are 24 molecular descriptors calculated by QikProp software, which are most relevant for the determination the #star parameter . For a given parameter, 1 #star corresponds to the computed property of a molecule falling outside the range for 95% of known drugs. A plot of the #stars parameter (on x-axis) against the corresponding counts (on y-axis) in the AfroDb is shown within the same set of axes with those of the “drug-like”, “lead-like”, and “fragment-like” standard subsets, (Figure 6). The criteria for the respective standard subsets were defined as (MW <500; log P<5; HBD ≤5; HBA ≤10) , (150≤ MW ≤350; log P≤4; HBD ≤3; HBA ≤6) – and (MW ≤250; −2≤ log P≤3; HBD <3; HBA <6; NRB <3) . It was observed that about 48% of the compounds within AfroDb showed #star = 0, while about 79% had #star ≤2. Among the 610 compounds of the “drug-like” subset, 73.4% had pharmacokinetic descriptors within the acceptable range for 95% of known drugs, while 98.9% showed #stars ≤2. The “lead-like” and “fragment-like” subsets were respectively 75.2% and 76.0% compliant for all of the 24 most relevant computed descriptors. The average values for 19 selected computed descriptors for all 4 compound libraries have been shown in Table 2. These values indicate a high probability of finding drug leads within the AfroDb compound library.
Blue = AfroDb library, red = “drug-like” subset, green = “lead-like” subset and violet = “fragment-like” subset.
Two processes determine the bioavailability of a compound; absorption and liver first-pass metabolism . The solubility and permeability of the compound, as well as interactions with transporters and metabolizing enzymes in the gut wall are factors responsible for absorption, while metabolism depends on the functional group types present. The computed parameters used to assess oral absorption revolved around Jorgensen’s famous “Rule of Three” (ro3). According to the ro3, a compliance with all or some of the rules (BIPcaco-2>22 nm/s and # primary metabolites <7) is indicative of likelihood to oral availability. Thus, the most important parameters often considered are the predicted aqueous solubility, , the conformation-independent predicted aqueous solubility, CI, the predicted qualitative human oral absorption, the predicted % human oral absorption. The solubility calculation procedure implemented depends on the similarity property space between the given molecule and its most similar analogue within the experimental training set used to develop the model implemented in QikProp, i.e., if the similarity is <0.9, then the QikProp predicted value is taken. Otherwise, the predicted property, , is adjusted such that:(1)where is the similarity, while and are the respective experimental and QikProp predictions for the most similar molecule within the training set. In equation (1), if then the predicted property is equal to the measured experimental property of the training set compound. The distribution curves for two of the three determinants for the ro3 (and BIPcaco-2) are shown in Figure S1. In general 43.6% of the AfroDb library was compliant with the ro3, while the respective % compliances for the various subsets were 67.6%, 92.3% and 100% for the “drug-like”, “lead-like” and “fragment-like” libraries. The most remarkable among the individual computed parameters was , which was met by 73.8% of the compounds within the AfroDb library. This property showed a Gaussian distribution for the “lead-like” and “fragment-like” subsets. The predicted apparent Caco-2 cell permeability, BIPcaco-2 (in nm s−1), model the permeability of the gut-blood barrier (for non-active transport). Even though the Caco-2 penetration parameter is not often correctly predicted computationally , only 38.2% of the compounds fell within the respected range for the BIPcaco-2 criterion. Histograms showing the predicted qualitative human oral absorption parameter (in the scale 1 = low, 2 = medium and 3 = high) are represented in Figure S2. It was observed that 55.8% of the compounds in AfroDb were predicted to have high human oral absorption. The predicted % human oral absorption (on 0 to 100% scale) shows a similar trend, 47.0% of the compounds being predicted to be absorbed at 100% and 63.1% of the compounds predicted to be absorbed at >90%.
The size of a molecule, as well as its capacity to make hydrogen bonds, its overall lipophilicity and its shape and flexibility are important properties to consider when determining permeability. Molecular flexibility has been seen as a parameter which is dependent on the NRB, a property which influences bioavailability in rats . As previously mentioned, the distribution of the NRB for this dataset showed a peak value lying between 3 and 4 RBs, with an average value of 6.3 (Table 2). The gap between the average and peak values could be explained by some degree of conformational flexibility in some of the bulkiest NPs (having as many as 61 RBs).
Prediction of Blood–brain Barrier (BBB) Penetration
The BBB partition coefficient is a good indicator of the ability of a drug to have access to the central nervous system (CNS). Drugs which are too polar do not cross the BBB. The blood/brain partition coefficients (logB/B) were computed and used as a predictor for access to the CNS. The predicted CNS activity was computed on a −2 (inactive) to +2 (active) scale and showed that only 3.8% of the compounds in AfroDb could be active in the CNS (predicted CNS activity >1). A distribution of logB/B (Figure 7A) showed a Gaussian-shaped curve with a peak at −0.5 logB/B units (the same for all the standard subsets), with >96% of the compounds in AfroDb falling within the recommended range for the predicted brain/blood partition coefficient (−3.0 to 1.2). Madin-Darby canine kidney (MDCK) monolayers, are widely used to make oral absorption estimates, the reason being that these cells also express transporter proteins, but only express very low levels of metabolizing enzymes . They are also used as an additional criterion to predict BBB penetration. Thus, our calculated apparent MDCK cell permeability could be considered to be a good mimic for the BBB penetration (for non-active transport). It was estimated that about 53% of the compounds had apparent MDCK cell permeability values falling within the recommended range of 25–500 nm s−1 for 95% of known drugs. This situation was not improved in the “drug-like” and “lead-like” subsets (∼55% and ∼44% respectively).
Prediction of Dermal Penetration
This factor is important for drugs administered through the skin. The distribution of computed skin permeability parameter, showed smooth Gaussian-shaped graphs centred at −2.5 units for all 4 datasets (Figure S3), with ∼92% of the compounds in the AfroDb database falling within the recommended range for >95% of known drugs. The predicted maximum transdermal transport rates, (in µ cm−2 hr−1), were computed from the predicted aqueous solubility () and skin permeability (), using the relation (2):(2)
This parameter showed variations from 0 to >1,600 µ cm−2 hr−1, with only about 2.5% of the compounds in AfroDb having predicted value of >100 µ cm−2 hr−1.
Prediction of Plasma-protein Binding
The degree to which a drug binds to the blood plasma proteins may seriously affect its efficacy. This is because binding of drugs to plasma proteins (like human serum albumin, lipoprotein, glycoprotein, α, β, and γ globulins) greatly reduces the quantity of the drug in general blood circulation and hence the less bound a drug is, the more efficiently it can traverse cell membranes or diffuse. The predicted plasma-protein binding has been estimated by the prediction of binding to human serum albumin; the parameter (recommended range is −1.5 to 1.5 for 95% of known drugs). Figure 7B shows the variation of this calculated parameter within the AfroDb dataset, as well as for the standard subsets. This equally gave smooth Gaussian-shaped curves centred on 0.5 units for the total and “drug-like” libraries and −0.5 units for the “lead-like” and “fragment-like” subsets. In addition, our calculations revealed that >92% of the compounds within the AfroDb library complied with this parameter, which is an indication that a bulk of the compounds are likely to circulate freely within the blood stream and hence have access to their respective target sites.
An estimated number of possible metabolic reactions has also been predicted by QikProp and used to determine whether the molecules can easily gain access to the target site after entering the blood stream. The average estimated number of possible metabolic reactions for the AfroDb library is between 6 and 7, while those of the standard subsets are respectively between 4 and 5, between 3 and 4 and between 2 and 3 (Table 2). Even though some of the compounds are likely to undergo as many as up to 25 metabolic reactions due to the complexity of some of the plant secondary metabolites within the database (Figure S4), up to about 80% of the compounds are predicted to undergo the recommended number of metabolic steps (1 to 8 reactions), with the situation improving to ∼91% and ∼97% in the “drug-like” and “lead-like” subsets respectively.
Prediction of Blockage of Human Ether-a-go-go-related Gene Potassium (HERG K+) Channel
Human ether-a-go-go related gene (HERG) encodes a potassium ion (K+) channel that is implicated in the fatal arrhythmia known as torsade de pointes or the long QT syndrome . The HERG K+ channel, which is best known for its contribution to the electrical activity of the heart that coordinates the heart’s beating, appears to be the molecular target responsible for the cardiac toxicity of a wide range of therapeutic drugs . HERG has also been associated with modulating the functions of some cells of the nervous system and with establishing and maintaining cancer-like features in leukemic cells . Thus, HERG K+ channel blockers are potentially toxic and the predicted IC50 values often provide reasonable predictions for cardiac toxicity of drugs in the early stages of drug discovery . In this work, the estimated or predicted IC50 values for blockage of this channel have been used to model the process in silico. The recommended range for predicted log IC50 values for blockage of HERG K+ channels (logHERG) is>−5. A distribution curve for the variation of the predicted logHERG is shown in Figure 7C, which is a left-slanted Gaussian-shaped curve, with a peak at −5.5 logHERG units for the total library and −4.5 logHERG units for all the standard subsets. It was observed that, in general, this parameter was predicted to fall within the recommended range for about 58% of the compounds within the AfroDb database, ∼70% for the “drug-like” subset and ∼90% for the “lead-like” subset.
In order to reduce redundancy and enhance the coverage of biological activity and chemical space, a dataset for virtual screening must have the requirement of diversity. In this case, we carried out a simple molecular descriptor comparison with a relatively larger known diverse library (the DIVERSet™ Database, containing 48,651 compounds) from the ChemBridge Corporation . Histograms showing the calculated descriptors; MW, HBA, HBD, log P, NRB,number of rings (NR), number of oxygens (NO), and total polar surface area (TPSA) are shown in Figure 8 for AfroDb (in light green) and the ChemBridge dataset (in red). The regions shown in dark green represent regions of intersection. The MW of the AfroDb dataset stretches up to about 1000 Da, while that of the ChemBridge dataset is restricted to the range 200≤ MW ≤500 Da. This observation could be explained by the complexity and large sizes of some of the compounds within the natural product library. The large proportion of very large and complex NPs in AfroDb, could also explain the average molar weight (Da), when compared to those of the standard “drug-like”, “lead-like” libraries and typical drugs (Da for typical drugs) . This same explanation holds for the trend which is observed in the distributions of log P, HBD, NCC, NO, NRB, NR, TPSA and HBA for AfroDb, when compared with the ChemBridge dataset. It was generally observed that the AfroDb dataset covers a different physico-chemical space from the ChemBridge Diversity dataset. Principal component analysis (PCA) was as well used as a means of comparing the extent of diversity of the two datasets. This consists in reducing the dimensionality of the calculated descriptors by linearly transforming the data, by calculating a new and smaller set of descriptors, which are uncorrelated and normalised (mean = 0, variance = 1). The PCA scatter plot of the previously calculated physico-chemical properties of the AfroDb (light green) and ChemBridge Diverset database (red), shown in Figure 9, is a visual representation of the molecules in the respective datasets, as described by the 3 selected principal components (PC1, PC2 and PC3). Each point shown corresponds to a molecule, the spread of the points representing the diversity of the respective datasets. The first three principal components (PCs) explain 80.1% (AfroDb) and 63.7% (ChemBridge) of the variance of the individual datasets. The larger number of outliers in the case of the AfroDb dataset (away for the centre and towards the sides of the cube) indicates a wider sampling of the chemical space compared to the ChemBridge Diverset collection.
Comparison of typical physico-chemical property distributions (MW, HBA, HBD, NCC, NO, NRB, log P, NR and TPSA) in the AfroDb (green) and ChemBridge Diverset (red) database. All histograms and scatterplots were generated with the R software .
Searching for Most Common Substructures
The most common substructure selection (MCSS) panel for compound selection (Figure 10) is based on substructures that can be synthetically combined and are common in “drug-like” molecules and allows a direct selection and identification of compounds containing such substructures. The panel highlights the large diversity of the rings present in the NPs of AfroDb.
Tautomer Generation and Synthetic Accessibility Evaluation
Of the 1008 compounds within AfroDb, we were able to identify 2308 possible tautomers, and respectively 1463, 524 and 103 possible tautomers for the standard subsets (“drug-like”, “lead-like” and “fragment-like”). The number of possible analogues were estimated at 9098, even the synthetic accessibility scores were quite weak for most of the proposed, as was expected for NPs.
Some Selected Highly Potent Compounds
Our discussion also includes selected compounds within the AfroDb database with promising biological activities, which have been isolated from African medicinal plants from the very first time. These have been summarised in Table 3, while their 2D structures are shown in Figure 11.
Among the promising compounds with anticancer activities, the trachylobane diterpene ent-trachyloban-3β-ol (04), derived from Croton zambesicus (Euphorbiaceae) in Benin, has been shown to exert a dose dependent cytotoxic effect, which varies between cell lines. Induction of apoptosis in HL-60 cells could be detected at a concentration of 50 µM after 24-h treatment. This compound was also able to induce apoptosis in human promyelocytic leukemia cells via caspase-3 activation in a concentration-dependent manner . Other promising cytotoxic agents included are six aporphines (05–10) from Cassytha filiformis, among which actinodaphnine (05), cassythine (06), and dicentrine (07) were also shown to possess antitrypanosomal properties in vitro on Trypanosoma brucei brucei . Hypoestoxide (HE, 11), a natural diterpenoid [a bicyclo (9, 3, 1) pentadecane], derived from Hypoestoes rosea growing in Nigeria  has been reported to be a potent nonsteroidal anti-inflammatory drug. This compound is also reported to be non mutagenic and possesses antiangiogenic and antitumor activities, also inhibiting the activity of IκB kinase , . Other potent anticancer compounds have been isolated from Gaillardia aristata growing in Egypt . These include neopulchellin (12) and 6α- hydroxyneopulchellin (13), with respective IC50 values of 0.43 and 0.32 mgmL-1 against human cancer cell lines (breast (MCF7)) and 0.46, 0.34 mgmL−1 against human colon (HCT116), respectively. Examples of NPs with anticancer properties derived from plants growing in the Southern African region include norviburtinal (14) and isopinnatal (15). These compounds were derived from Kigelia pinnata harvested in Zimbabwe and have shown cytotoxic activity against various cancer cell lines after 144 h exposure .
The anti-malarial properties of some of the compounds included in the AfroDb dataset have been discussed in several review papers –, as well as compounds with antiparasitic properties  and antimicrobial activities . One of the promising anti-malarial compounds include cryptolepine (16) derived from Sida acuta (Malvaceae), a plant growing in Côte d’Ivoire, which traditional healers commonly use for the treatment of malaria . This compound has shown antiplasmodial activity against the FcM29-Cameroon (chloroquine-resistant strain) and a Nigerian chloroquine-sensitive strain, with IC50 values of 0.13 and 0.17 µg/mL respectively.
Usefulness of the AfroDb Library
It is important to mention that virtual screening results sometimes provide insight and direct natural product chemists to search for theoretically active principles with attractive ADMET profiles, which have been previously isolated, but not tested for activity against specified drug targets (if samples are absent). This “resurrection” process could prove to be a better procedure for lead search than the random screening, which is common practice in most laboratories in Africa. AfroDb is constantly being updated; meanwhile a MySQL platform to facilitate the searching of this database and ordering of compound samples is under development within our group and will also be published subsequently. However, 3D structures of the compounds, as well as their physico-chemical properties that were used to evaluate “drug-likeness” and the DMPK profile, can be freely downloaded in the (Dataset S1). In addition, information about compound sample availability can be obtained on request from the authors of this paper or from the pan-African Natural Products Library (p-ANAPL) project –.
Materials and Methods
Data Sources and Cutoff Points for Biological Activities
The plant sources, geographical collection sites, chemical structures of pure compounds as well as their biological activities, were retrieved from literature sources comprising mainly published articles from across the major journals of natural product chemistry, as well as MSc and PhD theses, textbook chapters, as well as unpublished conference presentations (from personal communication with the authors). References were recorded from 1971 to 2013. Following criteria used by Mahmoudi et al.  and Wilcox et al. , a pure compound was considered highly active if IC50<0.06 µM, being active with 0.06 µM≤IC50≤5 µM, weakly active when 5 µM≤IC50≤10 µM and compounds with IC50>10 µM were considered inactive. Up to weakly active compounds were selected.
Generation of 3D Models, Optimization and Calculation of Molecular Descriptors
Based on the known chemical structures of the NPs, all 3D molecular structures were generated using the graphical user interface (GUI) of the MOE software  running on a Linux workstation with a 3.5 GHz Intel Core2 Duo processor. The 3D structures were generated using the builder module of MOE and energy minimization was subsequently carried out using the MMFF94 force field  until a gradient of 0.01 kcal/mol was reached. The 3D structures of the compounds were then saved as.mol2 files subsequently included into a MOE database (.mdb) file and converted to other file formats (.sdf,.mol,.mol2 and.ldb), which are suitable for use in several virtual screening workflow protocols. Up to 10 possible tautomers were generated per compound in the dataset. The MW, NRB, log P, HBA, HBD and Lipinski violations were calculated using the molecular descriptor calculator included in the QuSAR module of the MOE package .
Determination of ADMET Profiles
The previously prepared 1,008 low energy 3D chemical structures in the AfroDb library were saved in.mol2 format and initially treated with LigPrep , distributed by Schrodinger Inc. This implementation was carried out with the graphical user interface (GUI) of the Maestro software package , using the OPLS forcefield –. Protonation states at biologically relevant pH were correctly assigned (group I metals in simple salts were disconnected, strong acids were deprotonated, strong bases protonated, while topological duplicates and explicit hydrogens were added). A set of ADMET-related properties (a total of 46 molecular descriptors) were calculated by using the QikProp program  running in normal mode. QikProp generates physically relevant descriptors, and uses them to perform ADMET predictions. An overall ADME-compliance score – drug-likeness parameter (indicated by #stars), was used to assess the pharmacokinetic profiles of the compounds within the AfroDb library. The #stars parameter indicates the number of property descriptors computed by QikProp that fall outside the optimum range of values for 95% of known drugs. The methods implemented were developed by Jorgensen et al. – and among the calculated descriptors are: the total solvent-accessible molecular surface, in Å2 (probe radius 1.4 Å) (range for 95% of drugs: 300–1000 Å2); the hydrophobic portion of the solvent-accessible molecular surface, in Å2 (probe radius 1.4 Å) (range for 95% of drugs: 0–750 Å2); the total volume of molecule enclosed by solvent-accessible molecular surface, in Å3 (probe radius 1.4 Å) (range for 95% of drugs: 500–2000 Å3); the logarithm of aqueous solubility, (range for 95% of drugs: −6.0 to 0.5) , ; the logarithm of predicted binding constant to human serum albumin, (range for 95% of drugs: −1.5 to 1.2) ; the logarithm of predicted blood/brain barrier partition coefficient, logB/B (range for 95% of drugs: −3.0 to 1.0) –; the predicted apparent Caco-2 cell membrane permeability (BIPcaco-2) in Boehringer–Ingelheim scale, in nm/s (range for 95% of drugs: <5 low, >100 high) –; the predicted apparent Madin-Darby canine kidney (MDCK) cell permeability in nm s−1 (<25 poor, >500 great) ; the index of cohesion interaction in solids, Indcoh, calculated from the HBA, HBD and the surface area accessible to the solvent, SASA () by the relation Indcoh = HBAHBD1/2/ (0.0 to 0.05 for 95% of drugs) ; the globularity descriptor, Glob = , where r is the radius of the sphere whose volume is equal to the molecular volume (0.75 to 0.95 for 95% of drugs); the predicted polarizability, (13.0 to 70.0 for 95% of drugs); the predicted logarithm of IC50 value for blockage of HERG K+ channels, logHERG (concern<−5) –; the predicted skin permeability, (−8.0 to −1.0 for 95% of drugs) –; and the number of likely metabolic reactions, #metab (range for 95% of drugs: 1–8).
Diversity Analysis and Searching with Most Common Substructures
The ChemBridge Diverset dataset (48,651 compounds) was downloaded from the official ChemBridge webpage . The MW, NRB, log P, log S, HBA, HBD, THSA, TPSA, NO, NCC, NR and number of Lipinski violations were calculated using the molecular descriptor calculator included in the QuSAR module of the MOE package . The LibMCS program of JKlustor  was used for maximum common substructure clustering of the AfroDb database. In the MCSS search, only structures with MW ≤700 were included, since MCSS clustering is only feasible on small molecules. This means, only 975 of the 1008 compounds of the AfroDb were analyzed for MCSS. The compounds were fragmented using the RECAP algorithm  and the frequency of the generated fragments were analysed, leading to the identification of the most frequent substructures.
To the best of our knowledge, AfroDb represents the largest “drug-like” and diverse collection of 3D structures of NPs covering the geographical region of the entire African continent, which is readily available for download and use in virtual screening campaigns. Virtual screening workflows usually involve docking a compound library into the binding site of a target receptor and using scoring functions and binding free energy calculations to identify putative binders. The availability of 3D structures of the compounds to be used for docking is of utmost importance. Therefore the availability of such structures within AfroDb, as well as their calculated physico-chemical properties and indicators of “drug-likeness” within this newly developed database will facilitate the drug discovery process from leads that have been identified from African medicinal plants.
Distribution curves for compliance to Jorgensen’s “Rule of Three”. (A) calculated against count, (B) predicted BIPcaco-2 against count. Colour codes are as defined in Figure 6.
Histograms showing the distribution of human oral absorption predictions.
Distribution curves for the predicted skin penetration parameter. Colour codes are as defined in Figure 6.
Graphs showing the distribution of the predicted number of metabolic reactions for compounds in AfroDb. Colour codes are as defined in Figure 6.
SDF 3D structures of compounds currently included in AfroDb, along with computed physico-chemical descriptors used to predict drug-likeness and DMPK profile. This file is saved in.sdf format (which can be viewed using many drug discovery software including Maestro, MOE, Discovery studio, LigandScout, etc.) or into.ldb format using the software LigandScout. Requests for other file formats could be addressed to the authors of this article.
Mr. Bakoh Ndingkokhar, Chemistry Department, University of Buea, Cameroon assisted in the generation of the 3D models while Ms Inna Slynko, Department of Pharmaceutical Sciences, Martin-Luther University of Halle-Wittenberg, Germany, is acknowledged for assisting in carrying out the predicted MedChem Transformations and evaluation of the synthetic accessibilities.
Conceived and designed the experiments: WS LMM DZ FNK SMNE. Performed the experiments: FNK DZ MS SBB. Analyzed the data: MS FNK DZ SBB RM JAM. Contributed reagents/materials/analysis tools: WS DZ FNK LMM MS RM SMNE JAM LLL. Wrote the paper: FNK DZ SBB MS.
- 1. Akerele O (1993) In Summary of WHO guidelines for the assessment of herbal medicine. Herbalgram 28: 13–19.
- 2. Potterat O, Hamburger M (2008) Drug discovery and development with plant-derived compounds. In: Petersen F, Amstutz R, editors. Progress in drug research: natural compounds as drugs. Birhäusser: Verlag AG. Pp. 45–118.
- 3. Chin YW, Balunas MJ, Chai HB, Kinghorn AD (2006) Drug discovery from natural sources. The AAPS Journal 8(2): E239–E253.
- 4. Li JWH, Vederas JC (2009) Drug discovery and natural products: end of an era or an endless frontier? Science 325: 161–165.
- 5. Quinn RJ, Carroll AR, Pham MB, Baron P, Palframan ME, et al. (2008) Developing a drug-like natural product library. J Nat Prod 71: 464–468.
- 6. Harvey AL (2008) Natural products in drug discovery. Drug Discov Today 13: 894–901.
- 7. Newman DJ (2008) Natural products as leads to potential drugs: an old process or the new hope for drug discovery? J Med Chem 51: 2589–2599.
- 8. Efange SMN (2002) Natural products: a continuing source of inspiration for the medicinal chemist. In: Iwu MM, Wootton JC, editors. Advances in Phytomedicine. Amsterdam: Elsevier Science. 61–69.
- 9. Grabowski K, Baringhaus KH, Schneider G (2008) Scaffold diversity of natural products: inspiration for combinatorial library design. Nat Prod Rep 25: 892–904.
- 10. Grabowski K, Schneider G (2007) Properties and architecture of drugs and natural products revisited. Curr Chem Biol 1: 115–127.
- 11. Wetzel S, Schuffenhauer A, Roggo S, Ertl P, Waldmann H (2007) Cheminformatic analysis of natural products and their chemical space. Chimia Int J Chem 61: 355–360.
- 12. Hostettmann K, Marston A, Ndjoko K, Wolfender JL (2000) The potential of African plants as a source of drugs. Curr Org Chem 4: 973–1010.
- 13. Höltje HD, Sippl W, Rognan D, Folkers G (2008) Molecular modeling: basic principles and applications. Weinheim: WILEY-VCH Verlag GmbH & Co. KGaA. 181–258 p.
- 14. Kubinyi H (1998) Combinatorial and computational approaches in structure-based drug design. Curr. Opin. Drug Discov Devel 1, 16–27.
- 15. DiMasi JA, Hansen RW, Grabowsk HG (2003) The price of innovation: new estimates of drug development costs. J Health Econ 22: 151–185.
- 16. Ntie-Kang F, Mbah JA, Mbaze LM, Lifongo LL, Scharfe M, et al. (2013) CamMedNP: Building the Cameroonian 3D structural natural products database for virtual screening. BMC Complement Altern Med 13: 88.
- 17. Zofou D, Ntie-Kang F, Sippl W, Efange SMN (2013) Bioactive natural products derived from the Central African flora against Neglected Tropical Diseases and HIV. Nat Prod Rep 30: 1098–1120.
- 18. Ntie-Kang F, Onguéné PA, Scharfe M, Owono LCO, Megnassan E, et al.. (2013) ConMedNP: An in silico assessment of the “druglikeness” and medicinal potential of a natural product library from medicinal plants in Central Africa. RSC Advances, submitted.
- 19. Couzin J (1998) Supercomputing-computer experts urge new federal initiative. Science 281: 762.
- 20. Irwin JJ, Shoichet BK (2005) ZINC – A free database of commercially available compounds for virtual screening. J Chem Inf Model 45(1): 177–182.
- 21. Irwin JJ, Sterling T, Mysinger MM, Bolstad ES, Coleman RG (2012) ZINC: A free tool to discover chemistry for biology. J Chem Inf Model 52: 1757–1768.
- 22. Buzbee B (1993) Workstation clusters rise and shine. Science 261: 852–853.
- 23. Lipinski CA, Lombardo F, Dominy BW, Feeney PJ (1997) Experimental and computational approaches to estimate solubility and permeability in drug discovery and development settings. Adv Drug Delivery Rev 23: 3–25.
- 24. Chapman and Hall/CRC Press: Dictionary of Natural Products on CD-Rom. London; 2005.
- 25. ChemBridge Corporation website. Available: http://chembridge.com/Accessed 2013 June 13.
- 26. Feher M, Schmidt JM (2003) Property distributions: differences between drugs, natural products, and molecules from combinatorial chemistry. J Chem Inf Comput Sci 43: 218–227.
- 27. Nkanwen ERS, Gojayev AS, Wabo HK, Bankeu JJK, Iqbal MC, et al. (2013) Lanostane-type triterpenoid and steroid from the stem bark of Klainedoxa gabonensis. Fitoterapia 86: 108–114.
- 28. Mdee LK, Waibel R, Nkunya MHH, Jonker SA, Achenbach H (1998) Rosane diterpenes and bis-dinorditerpenes from Hugonia casteneifolia. Phytochemistry 49: 1107–1113.
- 29. Sob SVT, Wabo HK, Tang C-P, Tane P, Ngadjui BT, et al. (2011) Phenol esters and other constituents from the stem barks of Stereospermum acuminatissimum. J Asian Nat Prod Res 13: 1128–1134.
- 30. Schrödinger Press: QikProp 3.4 User Manual, LLC, New York, NY, 2011.
- 31. Teague SJ, Davis AM, Leeson PD, Opea TI (1999) The design of leadlike combinatorial libraries. Angew Chem, Int Ed 38: 3743–3748.
- 32. Oprea TI (2002) Current trends in lead discovery: are we looking for the appropriate properties? J Comput-Aided Mol Des 16: 325–334.
- 33. Schneider G (2002) Trends in virtual computational library design. Curr Med Chem 9: 2095–2102.
- 34. Verdonk ML, Cole JC, Hartshorn ML, Murray CW, Taylor RD (2003) Improved protein-ligand docking using GOLD. Proteins 52: 609–623.
- 35. Van de Waterbeemd H, Gifford E (2003) ADMET in silico modelling: towards prediction paradise? Nat Rev Drug Discov 2: 192–204.
- 36. Veber DF, Johnson SR, Cheng HY, Smith BR, Ward KW, et al. (2002) Molecular properties that influence the oral bioavailability of drug candidates. J Med Chem 45: 2615–2623.
- 37. Hedley PL, Jørgensen P, Schlamowitz S, Wangari R, Moolman-Smook J, et al. (2009) The genetic basis of long QT and short QT syndromes: A mutation update. Human Mutation 30: 1486–1511.
- 38. Vandenberg JI, Walker BD, Campbell TJ (2001) HERG K+ channels: Friend or foe. Trends Pharmacol Sci 22: 240–246.
- 39. Chiesa N, Rosati B, Arcangeli A, Olivotto M, Wanke E (1997) A novel role for HERG K+ channels: spike-frequency adaptation. J Physiol 501: 313–318.
- 40. Aronov AM (2005) Predictive in silico modeling for hERG channel blockers. Drug Discov Today 10: 149–155.
- 41. KhannaV, Ranganathan S (2011) Structural diversity of biologically interesting datasets: a scaffold analysis approach. J Cheminform 3: 30.
- 42. Block S, Gerkens P, Peulen O, Jolois O, Mingeot-Leclercq MP, et al. (2005) Induction of apoptosis in Human promyelocytic leukemia cells by a natural trachylobane diterpene. Anticancer Res 25: 363–368.
- 43. Hoet S, Stévigny C, Block S, Opperdoes F, Colson P, et al. (2004) Alkaloids from Cassytha filiformis and realted aporphines: antitrypanosomal activity, cytotoxicity, and interation with DNA and topoisomerases. Planta Med 70: 407–413.
- 44. Okogun JI, Adesomoju AA, Adesida GA, Lindner HJ, Habermehl G (1982) Roseanolone: A new diterpene from Hypoestoes rosea. Z Naturforsch 37c: 558–561.
- 45. Ojo-Amaize EA, Nchekwube EJ, Cottam HB, Bai R, Verdier-Pinard P, et al. (2002) Hypoestoxide, a natural nonmutagenic diterpenoid with antiangiogenic and antitumor activity: possible mechanisms of action. Cancer Res 62: 4007–4014.
- 46. Ojo-Amaize EA, Kapahi P, Kakkanaiah VN, Takahashi T, Shalom-Barak T, et al. (2001) Hypoestoxide, a novel anti-inflammatory natural diterpene, inhibits the activity of IκB kinase. Cellular Immunology 209: 149–157.
- 47. Salama MM, Kandil ZA, Islam WT (2012) Cytotoxic compounds from the leaves of Gaillardia aristata Pursh. growing in Egypt. Nat Prod Res 26 (22): 2057–2062.
- 48. Jackson SJ, Houghton PJ, Retsas S, Photiou A (2000) In vitro cytotoxicity of norviburtinal and isopinnatal from Kigelia pinnata against cancer cell lines. Planta Med 66: 758–761.
- 49. Schwikkard S, van Heerden RF (2002) Anti-malarial activity of plant metabolites. Nat Prod Rep 19(6): 675–692.
- 50. Magadula JJ, Erasto P (2009) Bioactive natural products derived from the East African flora. Nat Prod Rep 26: 1535–1554.
- 51. Titanji VPK, Zofou D, Ngemenya MN (2008) The anti-malarial potential of medicinal plants used for the treatment of malaria in Cameroonian Folk Medicine, Afr J Trad CAM. 5(3): 302–321.
- 52. Kuete V, Efferth T (2010) Cameroonian medicinal plants: pharmacology and derived natural products. Frontiers in Pharmacology 1: 123.
- 53. Ntie-Kang F, Lifongo LL, Mbaze LM, Ekwelle N, Owono Owono LC, et al. (2013) Cameroonian medicinal plants: a bioactivity versus ethnobotanical survey and chemotaxonomic classification. BMC Complement Altern Med 13(1): 147.
- 54. Hoet S, Opperdoes F, Brun R, Quetin-Leclercq J (2004) Natural products active against African trypanosomes: a step towards new drugs. Nat Prod Rep 21: 253–264.
- 55. Kuete V (2010) Potential of Cameroonian plants and derived products against microbial infections: a review. Planta Med 76: 1479–1491.
- 56. Banzouzi JT, Prado R, Menan H, Valentin A, Roumestan C, et al. (2004) Studies on medicinal plants of Ivory Coast: Investigation of Sida acuta for in vitro antiplasmodial activities and identification of an active constituent. Phytomedicine 11: 338–341.
- 57. Chibale K, Davies-Coleman M, Masimirembwa C (2012) Drug discovery in Africa: impacts of genomics, natural products, traditional medicines, insights into medicinal chemistry, and technology platforms in pursuit of new drugs. Springer.
- 58. pan-ANAPL: pan-African Natural Products Library. Available: http://www.linkedin.com/groups/pANPL-4098579/about/Accessed 2013 June 13.
- 59. Mahmoudi N, De Julian-Ortiz JV, Ciceron L, Galvez J, Mazier D, et al. (2006) Identification of new antimalarial drugs by linear discriminant analysis and topological virtual screening. J Antimicrob Chemother 57: 489–497.
- 60. Bickii J, Njifutie N, Foyere JA, Basco LK, Ringwald P (2000) In vitro antimalarial activity of limonoids from Khaya grandifoliola C.D.C. (Meliaceae). J Ethnopharmacol 69: 27–33.
- 61. Chemical Computing Group Inc.: Molecular Operating Environment Software. Montreal; 2010.
- 62. Halgren TA (1996) Merck molecular forcefield. J Comput Chem 17: 490–641.
- 63. Schrödinger: LigPrep software, version 2.5, LLC, New York, NY; 2011.
- 64. Schrödinger: Maestro, version 9.2, LLC, New York, NY; 2011.
- 65. Shivakumar D, Williams J, Wu Y, Damm W, Shelley J, et al. (2010) Prediction of absolute solvation free energies using molecular dynamics free energy perturbation and the OPLS force field. J Chem Theory Comput 6: 1509–1519.
- 66. Jorgensen WL, Maxwell DS, Tirado-Rives J (1996) Development and testing of the OPLS all-atom force field on conformational energetics and properties of organic liquids. J Am Chem Soc 118(45): 11225–11236.
- 67. Jorgensen WL, Tirado-Rives J (1988) The OPLS [optimized potentials for liquid simulations] potential functions for proteins, energy minimizations for crystals of cyclic peptides and crambin. J Am Chem Soc 110(6): 1657–1666.
- 68. Schrödinger: QikProp, version 3.4, LLC, New York, NY, 2011.
- 69. Jorgensen WL, Duffy EM (2002) Prediction of drug solubility from structure. Adv Drug Deliv Rev 54: 355–366.
- 70. Duffy EM, Jorgensen WL (2000) Prediction of properties from simulations: free energies of solvation in hexadecane, octanol, and water. J Am Chem Soc 122: 2878–2888.
- 71. Jorgensen WL, Duffy EM (2000) Prediction of drug solubility from Monte Carlo simulations. Bioorg Med Chem Lett 10: 1155–1158.
- 72. Colmenarejo G, Alvarez-Pedraglio A, Lavandera JL (2001) Cheminformatic models to predict binding affinities to human serum albumin. J Med Chem 44: 4370–4378.
- 73. Luco JM (1999) Prediction of brain-blood distribution of a large set of drugs from structurally derived descriptors using partial least squares (PLS) modelling. J Chem Inf Comput Sci 39: 396–404.
- 74. Kelder J, Grootenhuis PD, Bayada DM, Delbresine LP, Ploemen JP (1999) Polar molecular surface as a dominating determinant for oral absorption and brain pernetration of drugs. Pharm Res 16: 1514–1519.
- 75. Ajay, Bermis GW, Murkco MA (1999) Designing libraries with CNS activity. J Med Chem 42: 4942–4951.
- 76. Yazdanian M, Glynn SL, Wright JL, Hawi A (1998) Correlating partitioning and caco-2 cell permeability of structurally diverse small molecular weight compounds. Pharm Res 15: 1490–1494.
- 77. Irvine JD, Takahashi L, Lockhart K, Cheong J, Tolan JW, et al. (1999) MDCK (Madin-Darby canine kidney) cells: a tool for membrane permeability screening. J Pharm Sci 88: 28–33.
- 78. Stenberg P, Norinder U, Luthman K, Artursson P (2001) Experimental and computational screening models for the prediction of intestinal drug absorption. J Med Chem 44: 1927–1937.
- 79. Cavalli A, Poluzzi E, De Ponti F, Recanatini M (2002) Toward a pharmacophore for drugs inducing the long QT syndrome: Insights from a CoMFA Study of HERG K+ channel blockers. J Med Chem 45: 3844–3853.
- 80. De Ponti F, Poluzzi E, Montanaro N (2001) Organising evidence on QT prolongation and occurrence of Torsades de Pointes with non-antiarrhythmic drugs: a call for consensus. Eur J Clin Pharmacol 57: 185–209.
- 81. Potts RO, Guy RH (1992) Predicting skin permeability. Pharm Res 9: 663–669.
- 82. Potts RO, Guy RH (1995) A predictive algorithm for skin permeability: The effects of molecular size and hydrogen bond activity. Pharm Res 12: 1628–1633.
- 83. ChemAxon: JChem software, version 5.11.3; 2012. Website. Available: https://www.chemaxon.com/jchem/doc/user/LibMCS.html/Accessed 2013 June 13.
- 84. Lewell XQ, Judd DB, Watson SP, Hann MM (1998) RECAP-retrosynthetic combinatorial analysis procedure: a powerful new technique for identifying privileged molecular fragments with useful applications in combinatorial chemistry. J Chem Inf Comput Sci 38: 511–522.
- 85. R Core Team. R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna; 2012. Website http://www.R-project.org/Accessed 2013 June 13.