Markovian chemicals “in silico” design (MARCH-INSIDE), a promising approach for computer-aided molecular design III: 2.5D indices for the discovery of antibacterials

Machine Learning Study of Metabolic Networks vs ChEMBL Data of Antibacterial Compounds

Antibacterial drugs (AD) change the metabolic status of bacteria, contributing to bacterial death. However, antibiotic resistance and the emergence of multidrug-resistant bacteria increase interest in understanding metabolic network (MN) mutations and the interaction of AD vs MN. In this study, we employed the IFPTML = Information Fusion (IF) + Perturbation Theory (PT) + Machine Learning (ML) algorithm on a huge dataset from the ChEMBL database, which contains >155,000 AD assays vs >40 MNs of multiple bacteria species. We built a linear discriminant analysis (LDA) and 17 ML models centered on the linear index and based on atoms to predict antibacterial compounds. The IFPTML-LDA model presented the following results for the training subset: specificity (Sp) = 76% out of 70,000 cases, sensitivity (Sn) = 70%, and Accuracy (Acc) = 73%. The same model also presented the following results for the validation subsets: Sp = 76%, Sn = 70%, and Acc = 73.1%. Among the IFPTML nonlinear models, the k nearest neighbors (KNN) showed the best results with Sn = 99.2%, Sp = 95.5%, Acc = 97.4%, and Area Under Receiver Operating Characteristic (AUROC) = 0.998 in training sets. In the validation series, the Random Forest had the best results: Sn = 93.96% and Sp = 87.02% (AUROC = 0.945). The IFPTML linear and nonlinear models regarding the ADs vs MNs have good statistical parameters, and they could contribute toward finding new metabolic mutations in antibiotic resistance and reducing time/costs in antibacterial drug research.

Identification of Novel Antibacterials Using Machine Learning Techniques

Many pharmaceutical companies are avoiding the development of novel antibacterials due to a range of rational reasons and the high risk of failure. However, there is an urgent need for novel antibiotics especially against resistant bacterial strains. Available in silico models suffer from many drawbacks and, therefore, are not applicable for scoring novel molecules with high structural diversity by their antibacterial potency. Considering this, the overall aim of this study was to develop an efficient in silico model able to find compounds that have plenty of chances to exhibit antibacterial activity. Based on a proprietary screening campaign, we have accumulated a representative dataset of more than 140,000 molecules with antibacterial activity against Escherichia coli assessed in the same assay and under the same conditions. This intriguing set has no analogue in the scientific literature. We applied six in silico techniques to mine these data. For external validation, we used 5,000 compounds with low similarity towards training samples. The antibacterial activity of the selected molecules against E. coli was assessed using a comprehensive biological study. Kohonen-based nonlinear mapping was used for the first time and provided the best predictive power (av. 75.5%). Several compounds showed an outstanding antibacterial potency and were identified as translation machinery inhibitors in vitro and in vivo. For the best compounds, MIC and CC50 values were determined to allow us to estimate a selectivity index (SI). Many active compounds have a robust IP position.

Modeling Antibacterial Activity with Machine Learning and Fusion of Chemical Structure Information with Microorganism Metabolic Networks

Predicting the activity of new chemical compounds over pathogenic microorganisms with different metabolic reaction networks (MRN _s) is an important goal due to the different susceptibility to antibiotics. The ChEMBL database contains >160 000 outcomes of preclinical assays of antimicrobial activity for 55 931 compounds with >365 parameters of activity (MIC, IC₅₀, etc.) and >90 bacteria strains of >25 bacterial species. In addition, the Leong and Barabàsi data set includes >40 MRNs of microorganisms. However, there are no models able to predict antibacterial activity for multiple assays considering both drug and MRN structures at the same time. In this work, we combined perturbation theory, machine learning, and information fusion techniques to develop the first PTMLIF model. The best linear model found presented values of specificity = 90.31/90.40 and sensitivity = 88.14/88.07 in training/validation series. We carried out a comparison to nonlinear artificial neural network (ANN) techniques and previous models from the literature. Next, we illustrated the practical use of the model with an experimental case of study. We reported for the first time the isolation and characterization of terpenes from the plant Cissus incisa. The antibacterial activity of the terpenes was experimentally determined. The more active compounds were phytol and α-amyrin, with MIC = 100 μg/mL for Vancomycin-resistant Enterococcus faecium and Acinetobacter baumannii resistant to carbapenems. These compounds are already known from other sources. However, they have been isolated and evaluated for the first time here against several strains of multidrug-resistant bacteria including World Health Organization (WHO) priority pathogens. Last, we used the model to predict the activity of these compounds versus other microorganisms with different MRNs in order to find other potential targets.

A general ANN-based multitasking model for the discovery of potent and safer antibacterial agents

Bacteria have been one of the world's most dangerous and deadliest pathogens for mankind, nowadays giving rise to significant public health concerns. Given the prevalence of these microbial pathogens and their increasing resistance to existing antibiotics, there is a pressing need for new antibacterial drugs. However, development of a successful drug is a complex, costly, and time-consuming process. Quantitative Structure-Activity Relationships (QSAR)-based approaches are valuable tools for shortening the time of lead compound identification but also for focusing and limiting time-costly synthetic activities and in vitro/vivo evaluations. QSAR-based approaches, supported by powerful statistical techniques such as artificial neural networks (ANNs), have evolved to the point of integrating dissimilar types of chemical and biological data. This chapter reports an overview of the current research and potential applications of QSAR modeling tools toward the rational design of more efficient antibacterial agents. Particular emphasis is given to the setup of multitasking models along with ANNs aimed at jointly predicting different antibacterial activities and safety profiles of drugs/chemicals under diverse experimental conditions.

Poisson parameters of antimicrobial activity: a quantitative structure-activity approach

A contingency of observed antimicrobial activities measured for several compounds vs. a series of bacteria was analyzed. A factor analysis revealed the existence of a certain probability distribution function of the antimicrobial activity. A quantitative structure-activity relationship analysis for the overall antimicrobial ability was conducted using the population statistics associated with identified probability distribution function. The antimicrobial activity proved to follow the Poisson distribution if just one factor varies (such as chemical compound or bacteria). The Poisson parameter estimating antimicrobial effect, giving both mean and variance of the antimicrobial activity, was used to develop structure-activity models describing the effect of compounds on bacteria and fungi species. Two approaches were employed to obtain the models, and for every approach, a model was selected, further investigated and found to be statistically significant. The best predictive model for antimicrobial effect on bacteria and fungi species was identified using graphical representation of observed vs. calculated values as well as several predictive power parameters.

Design, synthesis and in vitro antitumour activity of new heteroaryl ethylenes

Almond and VolSurf + modelling procedures allowed the structural design of new di- and mono-heteroaryl-ethylenes. The structural modifications suggested by the molecular modelling were verified by the synthesis of the designed molecules and by the evaluation of their in vitro activities against two lung tumour cell lines, A549 and H226. 2-{(E)-2-[5'-(Dibutylamino)-2,2'-bithien-5-yl]vinyl}-1-methylquinolinium iodide exhibited in vitro antiproliferative activity two orders of magnitude higher than that of the most active compound previously synthesized in our laboratory.

First computational chemistry multi-target model for anti-Alzheimer, anti-parasitic, anti-fungi, and anti-bacterial activity of GSK-3 inhibitors in vitro, in vivo, and in different cellular lines

In the work described here, we developed the first multi-target quantitative structure-activity relationship (QSAR) model able to predict the results of 42 different experimental tests for GSK-3 inhibitors with heterogeneous structural patterns. GSK-3β inhibitors are interesting candidates for developing anti-Alzheimer compounds. GSK-3β are also of interest as anti-parasitic compounds active against Plasmodium falciparum, Trypanosoma brucei, and Leishmania donovani; the causative agents for Malaria, African Trypanosomiasis and Leishmaniosis. The MARCH-INSIDE technique was used to quickly calculate total and local polarizability, n-octanol/water partition coefficients, refractivity, van der Waals area and electronegativity values to 4,508 active/non-active compounds as well as the average values of these indexes for active compounds in 42 different biological assays. Both the individual molecular descriptors and the average values for each test were used as input for a linear discriminant analysis (LDA). We discovered a classification function which used in training series correctly classifies 873 out of 1,218 GSK-3 cases of inhibitors (97.4%) and 2,140 out of 2,163 cases of non-active compounds (86.1%) in the 42 different tests. In addition, the model correctly classifies 285 out of 406 GSK-3 inhibitors (96.3%) and 710 out of 721 cases of non-active compounds (85.4%) in external validation series. The result is important because, for the first time, we can use a single equation to predict the results of heterogeneous series of organic compounds in 42 different experimental tests instead of developing, validating, and using 42 different QSAR models. Lastly, a double ordinate Cartesian plot of cross-validated residuals (first ordinate), standard residuals (second ordinate), and leverages (abscissa) defined the domain of applicability of the model as a squared area within ± 2 band for residuals and a leverage threshold of h = 0.0044.

Scoring function for DNA-drug docking of anticancer and antiparasitic compounds based on spectral moments of 2D lattice graphs for molecular dynamics trajectories

We introduce here a new class of invariants for MD trajectories based on the spectral moments pi(k)(L) of the Markov matrix associated to lattice network-like (LN) graph representations of Molecular Dynamics (MD) trajectories. The procedure embeds the MD energy profiles on a 2D Cartesian coordinates system using simple heuristic rules. At the same time, we associate the LN with a Markov matrix that describes the probabilities of passing from one state to other in the new 2D space. We construct this type of LNs for 422 MD trajectories obtained in DNA-drug docking experiments of 57 furocoumarins. The combined use of psoralens+ultraviolet light (UVA) radiation is known as PUVA therapy. PUVA is effective in the treatment of skin diseases such as psoriasis and mycosis fungoides. PUVA is also useful to treat human platelet (PTL) concentrates in order to eliminate Leishmania spp. and Trypanosoma cruzi. Both are parasites that cause Leishmaniosis (a dangerous skin and visceral disease) and Chagas disease, respectively; and may circulate in blood products collected from infected donors. We included in this study both lineal (psoralens) and angular (angelicins) furocoumarins. In the study, we grouped the LNs on two sets; set1: DNA-drug complex MD trajectories for active compounds and set2: MD trajectories of non-active compounds or no-optimal MD trajectories of active compounds. We calculated the respective pi(k)(L) values for all these LNs and used them as inputs to train a new classifier that discriminate set1 from set2 cases. In training series the model correctly classifies 79 out of 80 (specificity=98.75%) set1 and 226 out of 238 (Sensitivity=94.96%) set2 trajectories. In independent validation series the model correctly classifies 26 out of 26 (specificity=100%) set1 and 75 out of 78 (sensitivity=96.15%) set2 trajectories. We propose this new model as a scoring function to guide DNA-docking studies in the drug design of new coumarins for anticancer or antiparasitic PUVA therapy.

Docking and 3D-QSAR studies of acetohydroxy acid synthase inhibitor sulfonylurea derivatives

Docking and three dimensional quantitative-structure activity relationship (3D-QSAR) studies were performed on acetohydroxy acid synthase (AHAS) inhibitor sulfonylurea analogues with potential herbicidal activity. The 3D-QSAR studies were carried out using shape, spatial and electronic descriptors along with a few structural parameters. Genetic function approximation (GFA) was used as the chemometric tool for this analysis. The whole data set (n = 45) was divided into a training set (75% of the data set) and a test set (remaining 25%) on the basis of the K-means clustering technique on a standardised topological, physicochemical and structural descriptor matrix. Models developed from the training set were used to predict the activity of the test set compounds. All models were validated internally, externally and using the Y-randomisation technique. Docking studies suggested that the molecules bind within a pocket of the enzyme formed by some important amino acid residues (Met351, Asp375, Arg377, Gly509, Met570 and Val571). In QSAR studies, molecular shape analysis showed that bulky substitution at the R(1) position may enhance AHAS inhibitory activity. Charged surface area descriptors suggested that negative charge distributed over a large surface area may enhance this activity. The hydrogen bond acceptor parameter supported the charged surface area descriptors and suggested that, for better activity, the number of electronegative atoms present in the molecule should be high. The spatial descriptors show that, for better activity, the molecules should possess a bulky substituent and a small substitution at the R(2) and R(3) positions, respectively.

Quantitative structure-activity relationship and complex network approach to monoamine oxidase A and B inhibitors

The work provides a new model for the prediction of the MAO-A and -B inhibitor activity by the use of combined complex networks and QSAR methodologies. On the basis of the obtained model, we prepared and assayed 33 coumarin derivatives, and the theoretical prediction was compared with the experimental activity data. The model correctly predicted 27 compounds, and most of the active derivatives showed IC 50 values in the muM-nM range against both the MAO-A and MAO-B isoforms. Compound 14 shows the same MAO-A inhibitory activity (IC 50 = 7.2 nM), as clorgyline used as a reference inhibitor and has the highest MAO-A specificity (1000-fold higher compared to MAO-B). On the other hand, compounds 24 (IC 50 = 1.2 nM) and 28 (IC 50 = 1.5 nM) show higher activity than selegiline (IC 50 = 19.6 nM) and high MAO-B selectivity with 100-fold and 1600-fold inhibition levels, with respect to the MAO-A isoform.

3D-MEDNEs: an alternative "in silico" technique for chemical research in toxicology. 2. quantitative proteome-toxicity relationships (QPTR) based on mass spectrum spiral entropy

Low range mass spectra (MS) characterization of serum proteome offers the best chance of discovering proteome-(early drug-induced cardiac toxicity) relationships, called here Pro-EDICToRs. However, due to the thousands of proteins involved, finding the single disease-related protein could be a hard task. The search for a model based on general MS patterns becomes a more realistic choice. In our previous work ( González-Díaz, H. , et al. Chem. Res. Toxicol. 2003, 16, 1318- 1327 ), we introduced the molecular structure information indices called 3D-Markovian electronic delocalization entropies (3D-MEDNEs). In this previous work, quantitative structure-toxicity relationship (QSTR) techniques allowed us to link 3D-MEDNEs with blood toxicological properties of drugs. In this second part, we extend 3D-MEDNEs to numerically encode biologically relevant information present in MS of the serum proteome for the first time. Using the same idea behind QSTR techniques, we can seek now by analogy a quantitative proteome-toxicity relationship (QPTR). The new QPTR models link MS 3D-MEDNEs with drug-induced toxicological properties from blood proteome information. We first generalized Randic's spiral graph and lattice networks of protein sequences to represent the MS of 62 serum proteome samples with more than 370 100 intensity ( I i ) signals with m/ z bandwidth above 700-12000 each. Next, we calculated the 3D-MEDNEs for each MS using the software MARCH-INSIDE. After that, we developed several QPTR models using different machine learning and MS representation algorithms to classify samples as control or positive Pro-EDICToRs samples. The best QPTR proposed showed accuracy values ranging from 83.8% to 87.1% and leave-one-out (LOO) predictive ability of 77.4-85.5%. This work demonstrated that the idea behind classic drug QSTR models may be extended to construct QPTRs with proteome MS data.

Computational chemistry approach for the early detection of drug-induced idiosyncratic liver toxicity

Idiosyncratic drug toxicity (IDT), considered as a toxic host-dependent event, with an apparent lack of dose response relationship, is usually not predictable from early phases of clinical trials, representing a particularly confounding complication in drug development. Albeit a rare event (usually <1/5000), IDT is often life threatening and is one of the major reasons new drugs never reach the market or are withdrawn post marketing. Computational methodologies, like the computer-based approach proposed in the present study, can play an important role in addressing IDT in early drug discovery. We report for the first time a systematic evaluation of classification models to predict idiosyncratic hepatotoxicity based on linear discriminant analysis (LDA), artificial neural networks (ANN), and machine learning algorithms (OneR) in conjunction with a 3D molecular structure representation and feature selection methods. These modeling techniques (LDA, feature selection to prevent over-fitting and multicollinearity, ANN to capture nonlinear relationships in the data, as well as the simple OneR classifier) were found to produce QSTR models with satisfactory internal cross-validation statistics and predictivity on an external subset of chemicals. More specifically, the models reached values of accuracy/sensitivity/specificity over 84%/78%/90%, respectively in the training series along with predictivity values ranging from ca. 78 to 86% of correctly classified drugs. An LDA-based desirability analysis was carried out in order to select the levels of the predictor variables needed to trigger the more desirable drug, i.e. the drug with lower potential for idiosyncratic hepatotoxicity. Finally, two external test sets were used to evaluate the ability of the models in discriminating toxic from nontoxic structurally and pharmacologically related drugs and the ability of the best model (LDA) in detecting potential idiosyncratic hepatotoxic drugs, respectively. The computational approach proposed here can be considered as a useful tool in early IDT prognosis.

Computational chemistry development of a unified free energy Markov model for the distribution of 1300 chemicals to 38 different environmental or biological systems

Predicting tissue and environmental distribution of chemicals is of major importance for environmental and life sciences. Most of the molecular descriptors used in computational prediction of chemicals partition behavior consider molecular structure but ignore the nature of the partition system. Consequently, computational models derived up-to-date are restricted to the specific system under study. Here, a free energy-based descriptor (DeltaG(k)) is introduced, which circumvent this problem. Based on DeltaG(k), we developed for the first time a single linear classification model to predict the partition behavior of a broad number of structurally diverse drugs and other chemicals (1300) for 38 different partition systems of biological and environmental significance. The model presented training/predicting set accuracies of 91.79/88.92%. Parametrical assumptions were checked. Desirability analysis was used to explore the levels of the predictors that produce the most desirable partition properties. Finally, inversion of the partition direction for each one of the 38 partition systems evidences that our models correctly classified 89.08% of compounds with an uncertainty of only +/-0.17% independently of the direction of the partition process used to seek the model. Other 10 different classification models (linear, neural networks, and genetic algorithms) were also tested for the same purposes. None of these computational models favorably compare with respect to the linear model indicating that our approach capture the main aspects that govern chemicals partition in different systems.

2D-RNA-coupling numbers: A new computational chemistry approach to link secondary structure topology with biological function

Methods for prediction of proteins, DNA, or RNA function and mapping it onto sequence often rely on bioinformatics alignment approach instead of chemical structure. Consequently, it is interesting to develop computational chemistry approaches based on molecular descriptors. In this sense, many researchers used sequence-coupling numbers and our group extended them to 2D proteins representations. However, no coupling numbers have been reported for 2D-RNA topology graphs, which are highly branched and contain useful information. Here, we use a computational chemistry scheme: (a) transforming sequences into RNA secondary structures, (b) defining and calculating new 2D-RNA-coupling numbers, (c) seek a structure-function model, and (d) map biological function onto the folded RNA. We studied as example 1-aminocyclopropane-1-carboxylic acid (ACC) oxidases known as ACO, which control fruit ripening having importance for biotechnology industry. First, we calculated tau(k)(2D-RNA) values to a set of 90-folded RNAs, including 28 transcripts of ACO and control sequences. Afterwards, we compared the classification performance of 10 different classifiers implemented in the software WEKA. In particular, the logistic equation ACO = 23.8 . tau(1)(2D-RNA) + 41.4 predicts ACOs with 98.9%, 98.0%, and 97.8% of accuracy in training, leave-one-out and 10-fold cross-validation, respectively. Afterwards, with this equation we predict ACO function to a sequence isolated in this work from Coffea arabica (GenBank accession DQ218452). The tau(1)(2D-RNA) also favorably compare with other descriptors. This equation allows us to map the codification of ACO activity on different mRNA topology features. The present computational-chemistry approach is general and could be extended to connect RNA secondary structure topology to other functions.

QSAR study of anticoccidial activity for diverse chemical compounds: Prediction and experimental assay of trans-2-(2-nitrovinyl)furan

In this work we report a QSAR model that discriminates between chemically heterogeneous classes of anticoccidial and non-anticoccidial compounds. For this purpose we used the Markovian Chemicals in silico Design (MARCH-INSIDE) approach J. Mol. Mod.2002, 8, 237-245; J. Mol. Mod.2003, 9, 395-407]. Linear discriminant analysis allowed us to fit the discriminant function. This function correctly classifies 86.67% of anticoccidial compounds and 96.23% of inactive compounds in the training series. Overall classification is 94.12%. We validated the model by means of an external predicting series, with 86.96% of global predictability. Remarkably, the present model is based on topological as well as configuration-dependent molecular descriptors. Therefore, the model performs timely calculations and allows discrimination between Z/E and chiral isomers. Finally, to exemplify the use of the model in practice we report the prediction and experimental assay of trans-2-(2-nitrovinyl)furan. It is notable that lesion control was 72.86% at mg/kg of body weight with respect to 60% at 125 mg/kg for amprolium (control drug). The back-projection map for this compound predicts a high level of importance for the double bond and for the nitro group in the trans position. We conclude that the MARCH-INSIDE approach enables the accurate fast track identification of anticoccidial hits. Moreover, trans-2-(2-nitrovinyl)furan seems to be a promising drug for the treatment of coccidiosis.

On an aspect of calculated molecular descriptors in QSAR studies of quinolone antibacterials

The re-emergence of tuberculosis infections, which are resistant to conventional drug therapy, has steadily risen in the last decade and as a result of that, fluoroquinolone drugs are being used as the second line of action. But there is hardly any study to examine specific structure activity relationships of quinolone antibacterials against mycobacteria. In this paper, an attempt has been made to establish a quantitative structure activity relationship modeling for a series of quinolone compounds against Mycobacterium fortuitum and Mycobacterium smegmatis. Due to lack of sufficient physicochemical data for the anti-mycobacterial compounds, it becomes very difficult to develop predictive methods based on experimental data. The present paper is an effort for the development of QSARs from the standpoint of physicochemical, constitutional, geometrical, electrostatic and topological indices. Molecular descriptors have been calculated solely from the chemical structure of N-1, C-7 and 8 substituted quinolone compounds and ridge regression models have been developed which can explain a better structure-activity relationship. Consideration of an intermolecular similarity analysis approach that led to a successful computer program development in PERL language has been used for comparing the influence of various molecular descriptors in different data subsets. The comparison of relative effectiveness of the calculated descriptors in our ridge regression model gives rise to some interesting results.

A QSAR Model for in Silico Screening of MAO-A Inhibitors. Prediction, Synthesis, and Biological Assay of Novel Coumarins

This work explores the potential of the MARCH-INSIDE methodology to seek a QSAR for MAO-A inhibitors from a heterogeneous series of compounds. A Markov model was used to quickly calculate the molecular electron delocalization, polarizability, refractivity, and n-octanol/water partition coefficients for a series of 1406 active/nonactive compounds. LDA was subsequently used to fit a classification function. The model showed 92.8% and 91.8% global accuracy and predictability in training and validation studies. This QSAR model was validated through a virtual screening of a series of coumarin derivatives. The 15 selected compounds were prepared and evaluated as in vitro MAO-A inhibitors. The theoretical prediction was compared with the experimental results and the model correctly predicted 13 compounds with only two mistakes on compounds with activities very close to the cutoff point established for the model. Consequently, this method represents a useful tool for the "in silico" screening of MAO-A inhibitors.

Proteins QSAR with Markov average electrostatic potentials

Classic physicochemical and topological indices have been largely used in small molecules QSAR but less in proteins QSAR. In this study, a Markov model is used to calculate, for the first time, average electrostatic potentials xik for an indirect interaction between aminoacids placed at topologic distances k within a given protein backbone. The short-term average stochastic potential xi1 for 53 Arc repressor mutants was used to model the effect of Alanine scanning on thermal stability. The Arc repressor is a model protein of relevance for biochemical studies on bioorganics and medicinal chemistry. A linear discriminant analysis model developed correctly classified 43 out of 53, 81.1% of proteins according to their thermal stability. More specifically, the model classified 20/28, 71.4% of proteins with near wild-type stability and 23/25, 92.0% of proteins with reduced stability. Moreover, predictability in cross-validation procedures was of 81.0%. Expansion of the electrostatic potential in the series xi0, xi1, xi2, and xi3, justified the use of the abrupt truncation approach, being the overall accuracy >70.0% for xi0 but equal for xi1, xi2, and xi3. The xi1 model compared favorably with respect to others based on D-Fire potential, surface area, volume, partition coefficient, and molar refractivity, with less than 77.0% of accuracy [Ramos de Armas, R.; González-Díaz, H.; Molina, R.; Uriarte, E. Protein Struct. Func. Bioinf.2004, 56, 715]. The xi1 model also has more tractable interpretation than others based on Markovian negentropies and stochastic moments. Finally, the model is notably simpler than the two models based on quadratic and linear indices. Both models, reported by Marrero-Ponce et al., use four-to-five time more descriptors. Introduction of average stochastic potentials may be useful for QSAR applications; having xik amenable physical interpretation and being very effective.