Designing of inhibitors against drug tolerant Mycobacterium tuberculosis (H37Rv)
© Singla et al.; licensee Chemistry Central Ltd. 2013
Received: 23 November 2012
Accepted: 25 February 2013
Published: 8 March 2013
Mycobacterium tuberculosis (M.tb) is the causative agent of tuberculosis, killing ~1.7 million people annually. The remarkable capacity of this pathogen to escape the host immune system for decades and then to cause active tuberculosis disease, makes M.tb a successful pathogen. Currently available anti-mycobacterial therapy has poor compliance due to requirement of prolonged treatment resulting in accelerated emergence of drug resistant strains. Hence, there is an urgent need to identify new chemical entities with novel mechanism of action and potent activity against the drug resistant strains.
This study describes novel computational models developed for predicting inhibitors against both replicative and non-replicative phase of drug-tolerant M.tb under carbon starvation stage. These models were trained on highly diverse dataset of 2135 compounds using four classes of binary fingerprint namely PubChem, MACCS, EState, SubStructure. We achieved the best performance Matthews correlation coefficient (MCC) of 0.45 using the model based on MACCS fingerprints for replicative phase inhibitor dataset. In case of non-replicative phase, Hybrid model based on PubChem, MACCS, EState, SubStructure fingerprints performed better with maximum MCC value of 0.28. In this study, we have shown that molecular weight, polar surface area and rotatable bond count of inhibitors (replicating and non-replicating phase) are significantly different from non-inhibitors. The fragment analysis suggests that substructures like hetero_N_nonbasic, heterocyclic, carboxylic_ester, and hetero_N_basic_no_H are predominant in replicating phase inhibitors while hetero_O, ketone, secondary_mixed_amine are preferred in the non-replicative phase inhibitors. It was observed that nitro, alkyne, and enamine are important for the molecules inhibiting bacilli residing in both the phases. In this study, we introduced a new algorithm based on Matthews correlation coefficient called MCCA for feature selection and found that this algorithm is better or comparable to frequency based approach.
In this study, we have developed computational models to predict phase specific inhibitors against drug resistant strains of M.tb grown under carbon starvation. Based on simple molecular properties, we have derived some rules, which would be useful in robust identification of tuberculosis inhibitors. Based on these observations, we have developed a webserver for predicting inhibitors against drug tolerant M.tb H37Rv available at http://crdd.osdd.net/oscadd/mdri/.
Tuberculosis (TB), a disease caused by M.tb kills around 1.7 million people every year despite the availability of effective chemotherapy for more than half a century . The antibiotic resistant strains of M.tb have arisen primarily due to poor compliance resulting from prolonged therapy . The emergence of multiple drug-resistant (MDR), extensive drug-resistant (XDR) strains, and its association with HIV has severely affected the fight against TB . Mathematical models have predicted that the MDR-TB and XDR-TB epidemics have the potential to further expand, thus threatening the success of TB control programs attained over last few decades [4–6].
In humans, the pathogenic cycle of TB consists of three phases : i) an active TB disease phase with actively replicating bacteria; ii) a latent phase wherein bacteria achieves a phenotypically distinct drug resistant state; and iii) a reactivation phase. The active TB disease phase is characterized by exponential increase of the pathogen, and latent phase is characterized by dormant phase in which pathogen remains metabolically quiescent and is not infectious. However, the reactivation phase is characterized by transition of latent infection into active TB disease. The reactivation of the disease occur in nearly 10% of patients with functional immune system and no separate dataset of inhibitors for this phase of pathogenic cycle is available. Therefore, in this study, we have used two phase inhibitors namely active and latent phase.
In past, researchers across the globe have deposited high throughput experimental data from M.tb growth inhibition assays. In PubChem, numerous datasets consisting of both the specific target based as well as cell-based inhibition assays are available. Utilizing these datasets, few computational models have been developed in past [8–11]. However, these studies are of little significance as they failed to contemplate the effect of potential hits on the drug-resistant M.tb strains grown under nutrient starvation condition. Furthermore, these studies does not distinguish the inhibitors based on their activity in different phase of TB. Therefore, it is important to develop new theoretical models for predicting inhibitors that would be effective against replicative as well as non-replicative drug-resistant M.tb and could potentially treat active TB patients as well as latently infected individuals.
Experimental techniques used in identification of inhibitors of M.tb growth are very expensive, time-consuming, tedious and requires sophisticated infrastructure (BSL-3) for mitigation of risk of infection. Thus, there is an urgent need to develop in-silico models for predicting inhibitors against drug-tolerant M.tb. In past, a number of target based models have been developed using QSAR and docking [12–16] for identification of novel inhibitors against M.tb. However, impermeability of chemical compounds to the mycobacterial cell wall hindered them to act as good lead molecules. To the best of our knowledge, no attempt has been made to develop prediction models against phase specific drug-tolerant M.tb.
Despite the enormous progress in computational and medicinal chemistry, only few webservers namely KiDoQ , GDoQ  and CDD  for predicting the efficacy of potential antimycobacterial drug like molecules are freely available to the scientific community. In order to assist researchers in discovering new chemical entity (NCE) against tuberculosis, a systematic algorithm has been developed to predict the inhibitors of replicative and non-replicative drug tolerant M.tb H37Rv.
The confirmatory screening in this assay resulted in 1277 active and 1017 inactive compounds against non-replicating M.tb. After removing the compounds containing salt/ions, we got a final dataset of 2135 compounds, out of which 1206 were identified as inhibitors and 929 were non-inhibitors.
This dataset involved screening of 2294 compounds from the BioAssay-488890 and identified 1453 inhibitors and 841 non-inhibitors for M.tb residing in replicating phase. After removing the salt/ions containing compounds, the final dataset was composed of total 2135 compounds of which 1355 compounds acted as replication mode inhibitor and rest were non-inhibitors.
The SMART pattern is the fragments present in compounds with undesirable effect reported in past and found to be responsible for toxicity or other side-effects. Therefore, it is important to search these reactive, non-advisable functional groups in the compounds with drug-like potential. In this study, we have used SMART filter web application (http://pasilla.health.unm.edu/tomcat/biocomp/smartsfilter) with Abbott ALARM , Glaxo  and Pfizer LINT  SMART filters. In this software, each compound was evaluated for potential to pass each particular filter. A molecule matching to this filter is classified into the failed category. On this basis, it will identify the number of compounds that pass or fail any of the implemented filters.
Substructure fragment analysis
where Nfragment_phase is the number of compounds containing the fragment in a M.tb phase inhibitor. Ntotal is the total number of compounds in that phase, Nfragment_total is the total number of compounds containing the fragment, and Nclass is the number of compounds in the M.tb phase inhibitor.
Since the pharmacophore represents the critical point present in chemical structure and take part in protein interaction, thus we have explored these features present in our datasets. The pharmacophore features were generated for the three first line (rifampicin, ethambutol and streptomycin) and four second line (ethionamide, cycloserine, kanamycin, amikacin) M.tb drugs using pharmagist software . These pharmacophores (named pharmacophore-1, pharmacophore-2) were then used to search similar compounds among the inhibitors of Rep_dataset, and NRep_dataset.
The PaDEL software has the capability of calculating 10 different types of fingerprints and 813 2D-3D descriptors . The binary fingerprints are easy to calculate, informative and interpretable, therefore we have used these in our datasets (see Data source section). The bit-string fingerprint is represented by 0’s or 1’s for the absence or presence of a particular fragment. In this study, we have used four different types of fingerprints.
It has been previously recognized that amongst the huge number of descriptors, only a few are relevant for efficient model building . It is well known that the computation time increases diagonally with addition of parameters. Furthermore, some descriptors that increases the noise level tremendously affect the model quality. Therefore, selection of highly relevant descriptors is a crucial step to reduce the noise level and to build a robust classification model. Therefore, we adopted multilayer techniques by 1) removing highly correlated descriptors (> = 0.8 to > =0.4), 2) MCC based selection of descriptors, 3) frequency based selection. For example, initially calculated 881 PubChem fingerprints calculating using PaDEL software were reduced to 597 after removing useless fingerprints, then to 247 by removing highly correlated descriptors at correlation cutoff 0.6.
SVM based classification models
We have used support vector machine (SVM) for discrimination between inhibitors and non-inhibitors of drug tolerant M.tb for both replicative and non-replicating phases. SVM can handle complex structural features based on the statistical and optimizations theory. In optimization process, the most important parameter is kernel function and is represented by t that varies from 0, 1, 2 corresponds to linear, polynomial, and radial basis function (RBF). The purpose of kernel function is to build a hyperplane that could separate two classes of data more accurately. For RBF kernel, the other parameter values are g, c, and j where c is used to trade-off between training error and margin, j is used to assign the cost, important in imbalance dataset and g is the gamma factor. In this study, we used SVMlight software package, which is freely available and can be downloaded from http://www.cs.cornell.edu/People/tj/svm_light/. The performance of models was optimized using a systematic variation of these different SVM parameters and kernels.
Evaluation of performance
To evaluate the performance of the prediction model, we adopted a five-fold cross validation approach. In this approach, the whole data was divided into five sets. Four sets were used in training and remaining 5th set was used for testing. This process was repeated five times such that each set comes in test set one time. If a particular compound was active and the prediction also envisage the same, then this was classified as true positive (TP); if actual was active and prediction was inactive, then it was false negative (FN); if actual was inactive and prediction was active, then its false positive (FP); and if actual is inactive and prediction is also inactive, then it’s true negative (TN) . Once the model was constructed fitness of the model was assessed using the commonly used statistical parameters . We have also created receiver operating curve (ROC) to evaluate the performance of models using threshold independent parameters. ROC plots with area under the curve were created using ROCR package in R.
This study is based on high-throughput screening data from PubChem BioAssay for identifying potential inhibitors against drug tolerant M.tb H37Rv (replicative phase and non-replicative phase).
Analysis of inhibitors and non-inhibitors
Mean (SD) of molecular descriptors from the M.tb datasets, compared actives and inactives
Rep_dataset vs. NRep_dataset
Based on these rules, we also tried to understand the behaviour of new class of anti-tuberculosis molecules and found that out of 7 replication mode inhibitors (PA-824, OPC-67683, TMC207, SQ109, Thioridazine, Lineziod, PNU-100480), on an average 3 (42.8%) molecules satisfied these rules. Similarly out of 4 inhibitors (PA-824, Thioridazine, Linezolid, Motifloxin), an average 2 (50%) compounds followed these rules. In order to further support these rules, we also analyzed 81 inhibitors (out of 177 because rest were complex form) of tuberculosis studied by Ballell et. al., and observed that 77.77% compounds fulfill the condition of molecular weight, 56.79% followed logP criteria, and 27.16% agreed with condition of rotatable bond count. There were only 19.75% active compounds which does not satisfy any of these rules while rest 80.25% were following one or more rules [Additional file 2: Table S1].
Validation of dataset
SMART filtering of the datasets
SMART filtering number of failures (%) using SMART filter website
TB drugs (13)
Pfizer LINT (%)
Abbott ALARM (%)
Substructure fragment analysis
Frequency of 20 representative substructure fragments in the Rep_dataset and NRep_dataset
Vinylogous_carbonyl or carboxyl_derivative
As shown in Table 3, the substructure patterns like nitro, alkyne, enamine were presented more frequently in case of inhibitors of both the Rep_dataset and NRep_dataset as compared to non-inhibitors. However, the patterns like amine, tertiary_carbon, alkylarylthioether and secondary_carbon are not preferred in any class of the inhibitors.
The PaDEL software used in this study calculates 881 PubChem, 166 MACCS, 79 EState, 307 SubStructure fingerprints and each corresponds to a specific substructure fragment. In this study, we have developed computational models on both the datasets using these fingerprints as described below.
Model based on NRep_dataset
Model based on binary fingerprints
Results of different binary fingerprints for NRep_dataset calculated from PaDEL software
Model based on features selected using MCC and frequency based algorithms
Results of different binary fingerprints for NRep_dataset on selected 15 descriptors calculated from PaDEL software
Frequency based descriptors
Hybrid (MCC + Frequency)
Model based on Rep_dataset
Model based on binary fingerprints
Results of different binary fingerprints for Rep_dataset calculated from PaDEL software
Model based on features selected using MCC and frequency based algorithms
Results of different binary fingerprints for Rep_dataset on selected 15 descriptors calculated from PaDEL software
Frequency based descriptors
Hybrid (MCC + Frequency)
In contrast to the general antibacterial rules or models, there is no report for phase specific rules and very limited efforts have been made to derive such ‘rules’ for tuberculosis [32–35]. Therefore, in the present study, we tried to generate new phase specific rules for better inhibitor predictions and drug development against M.tb. Our analysis suggested that simple molecular properties of chemical compounds like molecular weight, logP, polar surface area etc. were playing an important role in crossing the mycobacterium cell wall and its killing. Based on this study, we propose that some properties like molecular weight of compounds >300 Da for replication inhibitors and <380 Da for compounds inhibiting tuberculosis growth in non-replication mode. Based on this study, we derived some rules for identifying inhibitors against M.tb (for details see Results section). We have also shown that some substructure patterns like nitro, alkyne, enamine were dominating in inhibitor class of both phases. Similarly, the substructure like amine, tertiary_carbon, alkylarylthioether, secondary_carbon were not preferred in any of the growth phase inhibitors. This study demonstrated that molecules targeting the replicative and non-replicative phases have different chemical and molecular properties. These variations could arise from differences in the cellular metabolism and composition of cell wall of M.tb in these two phases of pathogenic cycles. We also observed that out of 7 drugs on an average 3 satisfied these criteria for replication inhibitors and out of 4 drugs known to be active in latent phase, ~2 also satisfied these rules implying the applicability of these modified rules for identifying anti-tuberculosis molecules. However this observation also suggests that there is an urgent requirement to increase the dataset of antitubercular drugs to further improve these rules. As suggested previously, identification of the undesirable fragment is important in early stages of drug discovery to reduce the time and cost involved in optimization process . Our SMART filtering results are similar to that of previous studies. The substructure patterns, identified in this work will be helpful for TB research community to design most potent inhibitory molecules against M.tb.
Additionally, four types of binary fingerprints were used to develop classification models using SVM based machine learning approach. We observed that the reduction of descriptors even at > =0.6 correlation cutoff, is sufficient to develop a robust classification model. As reported in different studies, we also observed that descriptors selection was playing an important role in efficient model building [17, 18]. In the present work, we have introduced a new algorithm named MCCA (Matthews Correlation Coefficient Algorithm) for selection of informative descriptors/fingerprints.
In past, different studies have been done to predict M.tb inhibitors. The Bayesian based classification model developed by Ekins et. al. has good predictive power value >0.7 (in-term of AUC) on independent dataset . In 2011, another Bayesian based model was developed to differentiate inhibitors under aerobic vs anaerobic condition . But the major limitation of previous studies was that these were not able to predict the replication/non-replication phase specific inhibitors of M.tb based on carbon starvation model. In 2010, A report by Gengenbacher et. al showed that the behaviour of drugs like steptomycin, rifampicin, isoniazid etc. was entirely different in replication, hypoxia induced drug tolerant and nutrient depleted models . Secondly, these models were based on the use of commercial softwares, hence limiting their accessibility. Similarly, in 2011, Periwal et. al. developed model on three dataset with maximum MCC value of 0.52. In 2012, a computational model was developed using large datasets obtained from high throughput screening based on whole cell screening using microdilution alamar blue assay and achieved maximum AUC value of 0.748 . Although, Periwal et. al. used the free softwares for model development but the non-availability of free software/webserver of these study restrict the use of their model by the scientific community. Considering these observations, we have developed a computational model that could discriminate the active compounds from inactive ones in both phases. Based on this study, we have also developed a user friendly, freely available webserver to search for new active molecules. We anticipate that these findings will provide insight that could be used in future to identify novel inhibitors effective against M.tb in either replicative or non-replicative phase.
In summary, we have identified some important substructures that are present in M.tb inhibitors. The SMART based filtering had identified 164 compounds from replicative inhibitors dataset and 180 compounds from non-replicative inhibitors dataset that passed all these three filters (see Results section) would be useful in future to reduce the effect of poor ADMET properties. These compounds would be useful in future for virtual screening and designing new inhibitors against M.tb. This study is implemented in the form of open source webserver to assist scientific researcher, and to boost up the drug discovery process against M.tb.
Web service to community
One of the main reasons of slow progress in Computer Aided Drug Designing (CADD) is the lack of freely available softwares and its implementation in user-friendly webservers. Most of these studies were focused on commercial softwares and hence their implementation is difficult. Our major emphasis is to help scientific community by developing freely accessible webserver/softwares based on our study. Thus, we have used both commercial as well as open source softwares in this study. Based on that, we have developed a webserver using SVM based classification model. Additionally, we have implemented the pharmagist software for identifying pharmacophore features similar to the first line and second line anti-mycobacterial drugs. Server has been developed under Linux environment using CGI-Perl scripts. In this web server, there are three options for molecule submission, 1) Draw structure using JME editor (http://www.molinspiration.com/jme/), 2) By pasting molecule in mol/mol2 file format, 3) By file upload. The results of prediction is provided in the tabular format with prediction class (inhibitor or non-inhibitor) of both phase as well as pharmacophore features similar to first line as well as second line M.tb drugs present or absent.
Quantitative structural activity relationship
SMiles ARbitrary target specification
Molecular libraries small molecule repository
Area under curve
Support vector machine.
The author's are thankful to the Council of Scientific and Industrial Research (CSIR) (project OSDD and GENESIS BSC0121), and Department of Biotechnology (DBT) (project BTISNET), for financial assistance.
- Wright A, Zignol M, Van Deun A, Falzon D, Gerdes SR, Feldman K, Hoffner S, Drobniewski F, Barrera L, van Soolingen D, Boulabhal F, Paramasivan CN, Kam KM, Mitarai S, Nunn P, Raviglione M: Epidemiology of antituberculosis drug resistance 2002–07: an updated analysis of the global project on anti-tuberculosis drug resistance surveillance. Lancet. 2009, 373: 1861-1873. 10.1016/S0140-6736(09)60331-7.View ArticleGoogle Scholar
- Ogbodo1 SO, Okeke AC, Ugwuoru CDC, Chukwurah EF: Possible alternatives to reduce antibiotic resistance. Life Sci Med Res. 2011, 2011: LSMR-24Google Scholar
- Kaneko T, Cooper C, Mdluli K: Challenges and opportunities in developing novel drugs for TB. Future Med Chem. 2011, 3: 1373-1400. 10.4155/fmc.11.115.View ArticleGoogle Scholar
- Dye C, Espinal MA: Will tuberculosis become resistant to all antibiotics?. Proc Biol Sci. 2001, 268: 45-52. 10.1098/rspb.2000.1328.View ArticleGoogle Scholar
- Dye C, Williams BG: Criteria for the control of drug-resistant tuberculosis. Proc Natl Acad Sci U S A. 2000, 97: 8180-8185. 10.1073/pnas.140102797.View ArticleGoogle Scholar
- Basu S, Andrews JR, Poolman EM, Gandhi NR, Shah NS, Moll A, Moodley P, Galvani AP, Friedland GH: Prevention of nosocomial transmission of extensively drug-resistant tuberculosis in rural South African district hospitals: an epidemiological modelling study. Lancet. 2007, 370: 1500-1507. 10.1016/S0140-6736(07)61636-5.View ArticleGoogle Scholar
- Talaat AM, Ward SK, Wu CW, Rondon E, Tavano C, Bannantine JP, Lyons R, Johnston SA: Mycobacterial bacilli are metabolically active during chronic tuberculosis in murine lungs: insights from genome-wide transcriptional profiling. J Bacteriol. 2007, 189: 4265-4274. 10.1128/JB.00011-07.View ArticleGoogle Scholar
- Ekins S, Kaneko T, Lipinski CA, Bradford J, Dole K, Spektor A, Gregory K, Blondeau D, Ernst S, Yang J, Goncharoff N, Hohman MM, Bunin BA: Analysis and hit filtering of a very large library of compounds screened against Mycobacterium tuberculosis. Mol Biosyst. 2010, 6: 2316-2324. 10.1039/c0mb00104j.View ArticleGoogle Scholar
- Ekins S, Freundlich JS: Validating new tuberculosis computational models with public whole cell screening aerobic activity datasets. Pharm Res. 2011, 28: 1859-1869. 10.1007/s11095-011-0413-x.View ArticleGoogle Scholar
- Periwal V, Rajappan JK, Jaleel AU, Scaria V: Predictive models for anti-tubercular molecules using machine learning on high-throughput biological screening datasets. BMC Res Notes. 2011, 4: 504-10.1186/1756-0500-4-504.View ArticleGoogle Scholar
- Prathipati P, Ma NL, Keller TH: Global Bayesian models for the prioritization of antitubercular agents. J Chem Inf Model. 2008, 48: 2362-2370. 10.1021/ci800143n.View ArticleGoogle Scholar
- Dwivedi N, Mishra BN, Katoch VM: 2D-QSAR model development and analysis on variant groups of anti-tuberculosis drugs. Bioinformation. 2011, 7: 82-90. 10.6026/97320630007082.View ArticleGoogle Scholar
- Sun J, Cai S, Mei H, Li J, Yan N, Wang Y: Docking and 3D QSAR study of thiourea analogs as potent inhibitors of influenza virus neuraminidase. J Mol Model. 2010, 16: 1809-1818. 10.1007/s00894-010-0685-9.View ArticleGoogle Scholar
- Pytela O, Klimesova V: Effect of substitution on the antimycobacterial activity of 2-(substituted benzyl)sulfanyl benzimidazoles, benzoxazoles, and benzothiazoles–a quantitative structure-activity relationship study. Chem Pharm Bull (Tokyo). 2011, 59: 179-184. 10.1248/cpb.59.179.View ArticleGoogle Scholar
- Ramesh KV, Purohit M, Mekhala K, Krishnan M, Wagle K, Deshmukh S: Modeling the interactions of herbal drugs to beta-ketoacyl ACP synthase of mycobacterium tuberculosis H37Rv. J Biomol Struct Dyn. 2008, 25: 481-493. 10.1080/07391102.2008.10507195.View ArticleGoogle Scholar
- Kumar M, Sharma S, Srinivasan A, Singh TP, Kaur P: Structure-based in-silico rational design of a selective peptide inhibitor for thymidine monophosphate kinase of mycobacterium tuberculosis. J Mol Model. 2011, 17: 1173-1182. 10.1007/s00894-010-0821-6.View ArticleGoogle Scholar
- Garg A, Tewari R, Raghava GP, KiDo Q: Using docking based energy scores to develop ligand based model for predicting antibacterials. BMC Bioinforma. 2010, 11: 125-10.1186/1471-2105-11-125.View ArticleGoogle Scholar
- Singla D, Anurag M, Dash D, Raghava GP: A web server for predicting inhibitors against bacterial target GlmU protein. BMC Pharmacol. 2011, 11: 5-View ArticleGoogle Scholar
- Ekins S, Bradford J, Dole K, Spektor A, Gregory K, Blondeau D, Hohman M, Bunin BA: A collaborative database and computational models for tuberculosis drug discovery. Mol Biosyst. 2010, 6: 840-851. 10.1039/b917766c.View ArticleGoogle Scholar
- Wang Y, Bolton E, Dracheva S, Karapetyan K, Shoemaker BA, Suzek TO, Wang J, Xiao J, Zhang J, Bryant SH: An overview of the PubChem BioAssay resource. Nucleic Acids Res. 2010, 38 (Database issue): D255-D266.View ArticleGoogle Scholar
- Wang Y, Xiao J, Suzek TO, Zhang J, Wang J, Bryant SH: PubChem: a public information system for analyzing bioactivities of small molecules. Nucleic Acids Res. 2009, 37 (Web Server issue): W623-W633.View ArticleGoogle Scholar
- Huth JR, Mendoza R, Olejniczak ET, Johnson RW, Cothron DA, Liu Y, Lerner CG, Chen J, Hajduk PJ: ALARM NMR: a rapid and robust experimental method to detect reactive false positives in biochemical screens. J Am Chem Soc. 2005, 127: 217-224. 10.1021/ja0455547.View ArticleGoogle Scholar
- Hann M, Hudson B, Lewell X, Lifely R, Miller L, Ramsden N: Strategic pooling of compounds for high-throughput screening. J Chem Inf Comput Sci. 1999, 39: 897-902. 10.1021/ci990423o.View ArticleGoogle Scholar
- Blake JF: Identification and evaluation of molecular properties related to preclinical optimization and clinical fate. Med Chem. 2005, 1: 649-655. 10.2174/157340605774598081.View ArticleGoogle Scholar
- Shen J, Cheng F, Xu Y, Li W, Tang Y: Estimation of ADME properties with substructure pattern recognition. J Chem Inf Model. 2010, 50: 1034-1041. 10.1021/ci100104j.View ArticleGoogle Scholar
- Cheng F, Yu Y, Shen J, Yang L, Li W, Liu G, Lee PW, Tang Y: Classification of cytochrome P450 inhibitors and noninhibitors using combined classifiers. J Chem Inf Model. 2011, 51: 996-1011.View ArticleGoogle Scholar
- Jensen BF, Vind C, Padkjaer SB, Brockhoff PB, Refsgaard HH: In silico prediction of cytochrome P450 2D6 and 3A4 inhibition using gaussian kernel weighted k-nearest neighbor and extended connectivity fingerprints, including structural fragment analysis of inhibitors versus noninhibitors. J Med Chem. 2007, 50: 501-511. 10.1021/jm060333s.View ArticleGoogle Scholar
- Schneidman-Duhovny D, Dror O, Inbar Y, Nussinov R, Wolfson HJ: Deterministic pharmacophore detection via multiple flexible alignment of drug-like molecules. J Comput Biol. 2008, 15: 737-754. 10.1089/cmb.2007.0130.View ArticleGoogle Scholar
- Yap CW: PaDEL-descriptor: an open source software to calculate molecular descriptors and fingerprints. J Comput Chem. 2011, 32: 1466-1474. 10.1002/jcc.21707.View ArticleGoogle Scholar
- Dudek AZ, Arodz T, Galvez J: Computational methods in developing quantitative structure-activity relationships (QSAR): a review. Comb Chem High Throughput Screen. 2006, 9: 213-228. 10.2174/138620706776055539.View ArticleGoogle Scholar
- Ballell L, Bates RH, Young RJ, Alvarez-Gomez D, Alvarez-Ruiz E, Barroso V, Blanco D, Crespo B, Escribano J, González RLS, Huss S, Santos-Villarejo A, Martín-Plaza JJ, Mendoza A, Rebollo-Lopez MJ, Remuiñan-Blanco M, Lavandera JL, Pérez-Herran E, Gamo-Benito FJ, García-Bustos JF, Barros D, Castro JP, Cammack N: Fueling open-source drug discovery: 177 small-molecule leads against tuberculosis. ChemMedChem. 2013, 8: 313-321. 10.1002/cmdc.201200428.View ArticleGoogle Scholar
- O’Shea R, Moser HE: Physicochemical properties of antibacterial compounds: implications for drug discovery. J Med Chem. 2008, 51: 2871-2878. 10.1021/jm700967e.View ArticleGoogle Scholar
- Charifson PS, Grillot AL, Grossman TH, Parsons JD, Badia M, Bellon S, Deininger DD, Drumm JE, Gross CH, LeTiran A, Liao Y, Mani N, Nicolau DP, Perola E, Ronkin S, Shannon D, Swenson LL, Tang Q, Tessier PR, Tian SK, Trudeau M, Wang T, Wei Y, Zhang H, Stamos D: Novel dual-targeting benzimidazole urea inhibitors of DNA gyrase and topoisomerase IV possessing potent antibacterial activity: intelligent design and evolution through the judicious use of structure-guided design and structure-activity relationships. J Med Chem. 2008, 51: 5243-5263. 10.1021/jm800318d.View ArticleGoogle Scholar
- Cronin MT, Aptula AO, Dearden JC, Duffy JC, Netzeva TI, Patel H, Rowe PH, Schultz TW, Worth AP, Voutzoulidis K, Schuurmann G: Structure-based classification of antibacterial activity. J Chem Inf Comput Sci. 2002, 42: 869-878. 10.1021/ci025501d.View ArticleGoogle Scholar
- Cecchetti V, Filipponi E, Fravolini A, Tabarrini O, Bonelli D, Clementi M, Cruciani G, Clementi S: Chemometric methodologies in a quantitative structure-activity relationship study: the antibacterial activity of 6-aminoquinolones. J Med Chem. 1997, 40: 1698-1706. 10.1021/jm960385p.View ArticleGoogle Scholar
- van de Waterbeemd H, Gifford E: ADMET in silico modelling: towards prediction paradise?. Nat Rev Drug Discov. 2003, 2: 192-204. 10.1038/nrd1032.View ArticleGoogle Scholar
- Gengenbacher M, Rao SPS, Pethe K, Dick T: Nutrient-starved, non-replicating mycobacterium tuberculosis requires respiration, ATP synthase and isocitrate lyase for maintenance of ATP homeostasis and viability. Microbiology. 2010, 156: 81-87. 10.1099/mic.0.033084-0.View ArticleGoogle Scholar
- Periwal V, Kishtapuram S: Open source drug discovery consortium and scaria, V: computational models for in-vitro anti-tubercular activity of molecules based on high-throughput chemical biology screening datasets. BMC Pharmacol. 2012, 12: 1-View ArticleGoogle Scholar
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.