Prediction of human intestinal absorption of drug compounds from molecular structure

Pharmacokinetic study with computational tools in the medicinal chemistry course

To improve the teaching-learning process in the Medicinal Chemistry course, new strategies have been incorporated into practical classes of this fundamental discipline of the pharmaceutical curriculum. Many changes and improvements have been made in the area of medicinal chemistry so far, and students should be prepared for these new approaches with the use of technological resources in this field. Practical activities using computational techniques have been directed to the evaluation of chemical and physicochemical properties that affect the pharmacokinetics of drugs. Their objectives were to allow students to know these tools, to learn how to access them, to search for the structures of drugs and to analyze results. To the best of our knowledge, this is the first study in Brazil to demonstrate the use of computational practices in teaching pharmacokinetics. Practical classes using Osiris and Molinspiration were attractive to students, who developed the activities easily and acquired better theoretical knowledge.

GPD: a graph pattern diffusion kernel for accurate graph classification with applications in cheminformatics

Graph data mining is an active research area. Graphs are general modeling tools to organize information from heterogeneous sources and have been applied in many scientific, engineering, and business fields. With the fast accumulation of graph data, building highly accurate predictive models for graph data emerges as a new challenge that has not been fully explored in the data mining community. In this paper, we demonstrate a novel technique called graph pattern diffusion (GPD) kernel. Our idea is to leverage existing frequent pattern discovery methods and to explore the application of kernel classifier (e.g., support vector machine) in building highly accurate graph classification. In our method, we first identify all frequent patterns from a graph database. We then map subgraphs to graphs in the graph database and use a process we call "pattern diffusion" to label nodes in the graphs. Finally, we designed a graph alignment algorithm to compute the inner product of two graphs. We have tested our algorithm using a number of chemical structure data. The experimental results demonstrate that our method is significantly better than competing methods such as those kernel functions based on paths, cycles, and subgraphs.

Prediction of human pharmacokinetics—evaluation of methods for prediction of hepatic metabolic clearance

Methods for prediction of hepatic clearance (CLH) in man have been evaluated. A physiologically-based in-vitro to in-vivo (PB-IVIV) method with human unbound fraction in blood (fu,bl) and hepatocyte intrinsic clearance (CLint)-data has a good rationale and appears to give the best predictions (maximum ∼2-fold errors; < 25% errors for half of CL-predictions; appropriate ranking). Inclusion of an empirical scaling factor is, however, needed, and reasons include the use of cryopreserved hepatocytes with low activity, and inappropriate CLint- and fu,bl-estimation methods. Thus, an improvement of this methodology is possible and required. Neglect of fu,bl or incorporation of incubation binding does not seem appropriate. When microsome CLint-data are used with this approach, the CLH is underpredicted by 5- to 9-fold on average, and a 106-fold underprediction (attrition potential) has been observed. The poor performance could probably be related to permeation, binding and low metabolic activity. Inclusion of scaling factors and neglect of fu,bl for basic and neutral compounds improve microsome predictions. The performance is, however, still not satisfactory. Allometry incorrectly assumes that the determinants for CLH relate to body weight and overpredicts human liver blood flow rate. Consequently, allometric methods have poor predictability. Simple allometry has an average overprediction potential, > 2-fold errors for ∼1/3 of predictions, and 140-fold underprediction to 5800-fold overprediction (potential safety risk) range. In-silico methodologies are available, but these need further development. Acceptable prediction errors for compounds with low and high CLH should be ∼50 and ∼10%, respectively. In conclusion, it is recommended that PB-IVIV with human hepatocyte CLint and fu,bl is applied and improved, limits for acceptable errors are decreased, and that animal CLH-studies and allometry are avoided.

In silico prediction of human oral absorption based on QSAR analyses of PAMPA permeability

The parallel artificial membrane permeation assay (PAMPA) was developed as a model for the prediction of transcellular permeation in the process of drug absorption. Our research group has measured the PAMPA permeability of peptide-related compounds, diverse drugs, and agrochemicals. This work led to a classical quantitative structure-activity relationship (QSAR) equation for PAMPA permeability coefficients of structurally diverse compounds based on simple physicochemical parameters such as lipophilicity at a particular pH (log P(oct) and |pKa-pH|), H-bond acceptor ability (SA(HA)), and H-bond donor ability (SA(HD)). Since the PAMPA permeability of lipophilic compounds decreased with their apparent lipophilicity due to the unstirred water layer (UWL) barrier on membrane surfaces and to membrane retention, a bilinear QSAR model was introduced to explain the permeability of a broader set of compounds using the same physicochemical parameters as those used for the linear model. We also compared PAMPA and Caco-2 cell permeability coefficients of compounds transported by various absorption mechanisms. The compounds were classified according to their absorption pathway (passively transported compounds, actively transported compounds, and compounds excreted by efflux systems) in the plot of Caco-2 vs. PAMPA permeability. Finally, based on the QSAR analyses of PAMPA permeability, an in silico prediction model of human oral absorption for possibly transported compounds was proposed, and the usefulness of the model was examined.

Ionization-specific analysis of human intestinal absorption

This study presents a mechanistic QSAR analysis of human intestinal absorption of drugs and drug-like compounds using a data set of 567 %HIA values. Experimental data represent passive diffusion across intestinal membranes, and are considered to be reasonably free of carrier-mediated transport or other unwanted effects. A nonlinear model was developed relating %HIA to physicochemical properties of drugs (lipophilicity, ionization, hydrogen bonding, and molecular size). The model describes ion-specific intestinal permeability of drugs by both transcellular and paracellular routes, and also accounts for unstirred water layer effects. The obtained model was validated on two external data sets consisting of in vivo human jejunal permeability coefficients (P(eff)) and absorption rate constants (K(a)). Validation results demonstrate good predictive power of the model (RMSE = 0.35-0.45 log units for log K(a) and log P(eff)). High prediction accuracy together with clear physicochemical interpretation (log P, pK(a)) makes this model particularly suitable for use in property-based drug design.

Graph wavelet alignment kernels for drug virtual screening

In this paper, we introduce a novel statistical modeling technique for target property prediction, with applications to virtual screening and drug design. In our method, we use graphs to model chemical structures and apply a wavelet analysis of graphs to summarize features capturing graph local topology. We design a novel graph kernel function to utilize the topology features to build predictive models for chemicals via Support Vector Machine classifier. We call the new graph kernel a graph wavelet-alignment kernel. We have evaluated the efficacy of the wavelet-alignment kernel using a set of chemical structure-activity prediction benchmarks. Our results indicate that the use of the kernel function yields performance profiles comparable to, and sometimes exceeding that of the existing state-of-the-art chemical classification approaches. In addition, our results also show that the use of wavelet functions significantly decreases the computational costs for graph kernel computation with more than ten fold speedup.

Recent advances in computer-aided drug design

Modern drug discovery is characterized by the production of vast quantities of compounds and the need to examine these huge libraries in short periods of time. The need to store, manage and analyze these rapidly increasing resources has given rise to the field known as computer-aided drug design (CADD). CADD represents computational methods and resources that are used to facilitate the design and discovery of new therapeutic solutions. Digital repositories, containing detailed information on drugs and other useful compounds, are goldmines for the study of chemical reactions capabilities. Design libraries, with the potential to generate molecular variants in their entirety, allow the selection and sampling of chemical compounds with diverse characteristics. Fold recognition, for studying sequence-structure homology between protein sequences and structures, are helpful for inferring binding sites and molecular functions. Virtual screening, the in silico analog of high-throughput screening, offers great promise for systematic evaluation of huge chemical libraries to identify potential lead candidates that can be synthesized and tested. In this article, we present an overview of the most important data sources and computational methods for the discovery of new molecular entities. The workflow of the entire virtual screening campaign is discussed, from data collection through to post-screening analysis.

Prediction of human intestinal absorption by GA feature selection and support vector machine regression

QSAR (Quantitative Structure Activity Relationships) models for the prediction of human intestinal absorption (HIA) were built with molecular descriptors calculated by ADRIANA.Code, Cerius² and a combination of them. A dataset of 552 compounds covering a wide range of current drugs with experimental HIA values was investigated. A Genetic Algorithm feature selection method was applied to select proper descriptors. A Kohonen's self-organizing Neural Network (KohNN) map was used to split the whole dataset into a training set including 380 compounds and a test set consisting of 172 compounds. First, the six selected descriptors from ADRIANA.Code and the six selected descriptors from Cerius² were used as the input descriptors for building quantitative models using Partial Least Square (PLS) analysis and Support Vector Machine (SVM) Regression. Then, another two models were built based on nine descriptors selected by a combination of ADRIANA.Code and Cerius² descriptors using PLS and SVM, respectively. For the three SVM models, correlation coefficients (r) of 0.87, 0.89 and 0.88 were achieved; and standard deviations (s) of 10.98, 9.72 and 9.14 were obtained for the test set.

Prediction of Permeability Coefficients of Compounds Through Caco-2 Cell Monolayer Using Artificial Neural Network Analysis

Artificial neural network (ANN) analysis was used to predict the permeability of selected compounds through Caco-2 cell monolayers. Previously reported models, which were shown to be useful in the prediction of permeability values, use many structural parameters. More complex equations have also been proposed using both linear and non-linear relationships, including ANN analysis and various structural parameters. But proposed models still need to be developed using different neuron patterns for more precise predictions and a better understanding of which factors affect the permeation. To develop a simple and useful model or method for easy prediction is also a general need. Permeability coefficients (log kp) were obtained from various literature sources. Some structural parameters were calculated using computer programs. Multiple linear regression analysis (MLRA) was used to predict Caco-2 cell permeability for the set of 50 compounds (r2=0.403). A successful ANN model was developed, and the ANN produced log kp values that correlated well with the experimental ones (r2=0.952). The permeability of a compound, famotidine, which has not previously been studied, through the Caco-2 cell monolayer was investigated, and its permeability coefficient determined. It was then possible to compare the experimental data with that predicted using the trained ANN with previously determined Caco-2 cell permeability values and structural parameters of compounds. The model was also tested using literature values. The developed and described ANN model in this publication does not require any experimental parameters; it could potentially provide useful and precise prediction of permeability for new drugs or other penetrants.

Computational model for predicting chemical substituent effects on passive drug permeability across parallel artificial membranes

Drug permeability is often a limiting step in drug action, requiring chemical optimization of a drug candidate to improve this property. Such optimization is typically performed in the context of a congeneric series, where substituents are varied to optimize the target property. Motivated by this need the present work examines the influence of chemical substituents on passive permeability (log P pass) across parallel artificial membranes (PAMPA) undertaken for three congeneric series of compounds; benzoic acids, pyridines and quinolines. PAMPA showed pyridine and quinoline to have high permeability and chemical substituents to typically reduce the permeability. On the contrary, benzoic acid showed poor permeability and chemical substituents typically increased the permeability. To quantitate these effects with respect to physical properties, models were built to explain and predict the permeability of these classes of compounds based on computed molecular descriptors. Models for the benzoic acid series in the ionized state indicated the solvent accessible surface area, cavity dispersion and the free energy of solvation in hexane as well as in water to dominate permeability. However, when the acid group is treated as neutral, the free energy of solvation in water, the fraction polar surface area, the polar surface area and difference in the free energy of solvation between hexane and water were important; these terms, among others, were also important for the neutral pyridine-quinoline series. Considering that the permeability of the benzoic acid series is about 2 orders of magnitude lower than the pyridines and quinolines and that a shift of approximately two pH units in the p K a of the acid group of benzoic acid will allow for the neutral species of the molecule to dominate under experimental conditions (pH = 6.5), these results suggest that the additional energy barrier associated with permeation of the benzoic acid series is associated with the need to protonate the acidic group, thereby forming the neutral species which may then cross the hydrophobic region of the membrane.

Synthesis, antibacterial and antifungal activities of some N-nitroso-2,6-diarylpiperidin-4-one semicarbazones and QSAR analysis

A series of N-nitroso-2,6-diarylpiperidin-4-one semicarbazones and thiosemicarbazones were synthesized, characterized by IR, NMR and elemental analysis. All the compounds were screened for their antibacterial activity against Gram-positive bacteria Bacillus subtilis, Staphylococcus aureus and Gram-negative bacteria Escherichia coli and fungi Candida albicans. These compounds have showed moderate and very good antibacterial activity. Quantitative Structure Activity Relationship (QSAR) analysis was performed for these compounds by the application of Semiempirical calculations and molecular modeling. Different logP values were also evaluated to further the analysis.

Structure-activity models of oral clearance, cytotoxicity, and LD50: a screen for promising anticancer compounds

BACKGROUND

Quantitative structure-activity relationship (QSAR) models have become popular tools to help identify promising lead compounds in anticancer drug development. Few QSAR studies have investigated multitask learning, however. Multitask learning is an approach that allows distinct but related data sets to be used in training. In this paper, a suite of three QSAR models is developed to identify compounds that are likely to (a) exhibit cytotoxic behavior against cancer cells, (b) exhibit high rat LD50 values (low systemic toxicity), and (c) exhibit low to modest human oral clearance (favorable pharmacokinetic characteristics). Models were constructed using Kernel Multitask Latent Analysis (KMLA), an approach that can effectively handle a large number of correlated data features, nonlinear relationships between features and responses, and multitask learning. Multitask learning is particularly useful when the number of available training records is small relative to the number of features, as was the case with the oral clearance data.

RESULTS

Multitask learning modestly but significantly improved the classification precision for the oral clearance model. For the cytotoxicity model, which was constructed using a large number of records, multitask learning did not affect precision but did reduce computation time. The models developed here were used to predict activities for 115,000 natural compounds. Hundreds of natural compounds, particularly in the anthraquinone and flavonoids groups, were predicted to be cytotoxic, have high LD50 values, and have low to moderate oral clearance.

CONCLUSION

Multitask learning can be useful in some QSAR models. A suite of QSAR models was constructed and used to screen a large drug library for compounds likely to be cytotoxic to multiple cancer cell lines in vitro, have low systemic toxicity in rats, and have favorable pharmacokinetic properties in humans.

Estimation of ADME Properties in Drug Discovery: Predicting Caco-2 Cell Permeability Using Atom-Based Stochastic and Non-stochastic Linear Indices

The in vitro determination of the permeability through cultured Caco-2 cells is the most often-used in vitro model for drug absorption. In this report, we use the largest data set of measured P(Caco-2), consisting of 157 structurally diverse compounds. Linear discriminant analysis (LDA) was used to obtain quantitative models that discriminate higher absorption compounds from those with moderate-poorer absorption. The best LDA model has an accuracy of 90.58% and 84.21% for training and test set. The percentage of good correlation, in the virtual screening of 241 drugs with the reported values of the percentage of human intestinal absorption (HIA), was greater than 81%. In addition, multiple linear regression models were developed to predict Caco-2 permeability with determination coefficients of 0.71 and 0.72. Our method compares favorably with other approaches implemented in the Dragon software, as well as other methods from the international literature. These results suggest that the proposed method is a good tool for studying the oral absorption of drug candidates.

CHEMICAL COMPOUND CLASSIFICATION WITH AUTOMATICALLY MINED STRUCTURE PATTERNS

In this paper we propose new methods of chemical structure classification based on the integration of graph database mining from data mining and graph kernel functions from machine learning. In our method, we first identify a set of general graph patterns in chemical structure data. These patterns are then used to augment a graph kernel function that calculates the pairwise similarity between molecules. The obtained similarity matrix is used as input to classify chemical compounds via a kernel machines such as the support vector machine (SVM). Our results indicate that the use of a pattern-based approach to graph similarity yields performance profiles comparable to, and sometimes exceeding that of the existing state-of-the-art approaches. In addition, the identification of highly discriminative patterns for activity classification provides evidence that our methods can make generalizations about a compound's function given its chemical structure. While we evaluated our methods on molecular structures, these methods are designed to operate on general graph data and hence could easily be applied to other domains in bioinformatics.

Modulation of Ganciclovir Intestinal Absorption in Presence of Absorption Enhancers

The purpose of this investigation was to study the influences of absorption enhancers in increasing oral bioavailability of Ganciclovir (GAN) by assessing the transepithelial permeation across cell monolayers in vitro and bioavailability in rats in vivo. The permeation of GAN across Caco-2 and MDCK cell monolayers in the absence/presence of dimethyl-beta-cyclodextrin (DMbetaCD), chitosan hydrochloride (CH), sodium lauryl sulphate (SLS), and their combinations was studied for a 2-h period. GAN was administered to rats in absence/presence of absorption enhancers and drug contents in plasma were estimated. We found that the apparent permeability coefficient (Papp) of GAN in absence of absorption enhancers (control) were 0.261 +/- 0.072 x 10(-6) and 0.486 +/- 0.063 x 10(-6) cm/s in Caco-2 and MDCK cell monolayers, respectively, whereas in the presence of DMbetaCD, CH, SLS, and their combinations, Papp of GAN increased by 5- to 25-fold and 7- to 33-fold as compared to control in Caco-2 and MDCK cell monolayers, respectively. However, in rats, the maximum enhancement in bioavailability of GAN during coadministration of these absorption enhancers was only fivefold compared to GAN control. To conclude, the absorption enhancers-DMbetaCD, CH, SLS, and their combinations demonstrated significant improvement in transepithelial permeation and bioavailability of GAN.

Artificial neural network models for prediction of intestinal permeability of oligopeptides

BACKGROUND

Oral delivery is a highly desirable property for candidate drugs under development. Computational modeling could provide a quick and inexpensive way to assess the intestinal permeability of a molecule. Although there have been several studies aimed at predicting the intestinal absorption of chemical compounds, there have been no attempts to predict intestinal permeability on the basis of peptide sequence information. To develop models for predicting the intestinal permeability of peptides, we adopted an artificial neural network as a machine-learning algorithm. The positive control data consisted of intestinal barrier-permeable peptides obtained by the peroral phage display technique, and the negative control data were prepared from random sequences.

RESULTS

The capacity of our models to make appropriate predictions was validated by statistical indicators including sensitivity, specificity, enrichment curve, and the area under the receiver operating characteristic (ROC) curve (the ROC score). The training and test set statistics indicated that our models were of strikingly good quality and could discriminate between permeable and random sequences with a high level of confidence.

CONCLUSION

We developed artificial neural network models to predict the intestinal permeabilities of oligopeptides on the basis of peptide sequence information. Both binary and VHSE (principal components score Vectors of Hydrophobic, Steric and Electronic properties) descriptors produced statistically significant training models; the models with simple neural network architectures showed slightly greater predictive power than those with complex ones. We anticipate that our models will be applicable to the selection of intestinal barrier-permeable peptides for generating peptide drugs or peptidomimetics.

ADME evaluation in drug discovery. 6. Can oral bioavailability in humans be effectively predicted by simple molecular property-based rules?

A critically evaluated database of human oral bioavailability for 768 chemical compounds is described in this study (http://modem.ucsd.edu/adme), which provides the scientific community a publicly available and reliable source for developing predictive models of human oral bioavailability. The correlations between several important molecular properties and human oral bioavailability were investigated and compared with an earlier report by analyzing the rat oral bioavailability data (J. Med. Chem. 2002, 45, 2615). We showed that the percentages of compounds meeting the criteria based on molecular properties does not distinguish compounds with poor oral bioavailability from those with acceptable values, which may suggest that no simple rule based on molecular properties can be used as general filters to predict oral bioavailability with high confidence. A data set of intestinal absorption was also examined and compared with that of oral bioavailability. The performance of these rules based on molecular properties in the prediction of intestinal absorption is obviously much better than that of oral bioavailability in term of false positive rate, and, therefore, the applications of the "rule-based" approaches on the prediction of human bioavailability should be very cautious.

ADME evaluation in drug discovery. 7. Prediction of oral absorption by correlation and classification

A critically evaluated database of human intestinal absorption for 648 chemical compounds is reported in this study, among which 579 are believed to be transported by passive diffusion. The correlation analysis between the intestinal absorption and several important molecular properties demonstrated that no single molecular property could be used as a good discriminator to efficiently distinguish the poorly absorbed compounds from those that are well absorbed. The theoretical correlation models for a training set of 455 compounds were proposed by using the genetic function approximation technique. The best prediction model contains four molecular descriptors: topological polar surface area, the predicted distribution coefficient at pH = 6.5, the number of violations of the Lipinski's rule-of-five, and the square of the number of hydrogen-bond donors. The model was able to predict the fractional absorption with an r = 0.84 and a prediction error (absolute mean error) of 11.2% for the training set. Moreover, it achieves an r = 0.90 and a prediction error of 7.8% for a 98-compound test set. The recursive partitioning technique was applied to find the simple hierarchical rules to classify the compounds into poor (%FA < or = 30%) and good (%FA > 30%) intestinal absorption classes. The high quality of the classification model was validated by the satisfactory predictions on the training set (correctly identifying 95.9% of the compounds in the poor-absorption class and 96.1% of the compounds in the good-absorption class) and on the test set (correctly identifying 100% of the compounds in the poor-absorption class and 96.8% of the compounds in the good-absorption class). We expect that, in the future, the rules for the prediction of carrier-mediated transporting and first pass metabolism can be integrated into the current hierarchical rules, and the classification model may become more powerful in the prediction of intestinal absorption or even human bioavailability. The databases of human intestinal absorption reported here are available for download from the supporting Web site: http://modem.ucsd.edu/adme.

Caco-2 cell permeability modelling: a neural network coupled genetic algorithm approach

The ability to cross the intestinal cell membrane is a fundamental prerequisite of a drug compound. However, the experimental measurement of such an important property is a costly and highly time consuming step of the drug development process because it is necessary to synthesize the compound first. Therefore, in silico modelling of intestinal absorption, which can be carried out at very early stages of drug design, is an appealing alternative procedure which is based mainly on multivariate statistical analysis such as partial least squares (PLS) and neural networks (NN). Our implementation of neural network models for the prediction of intestinal absorption is based on the correlation of Caco-2 cell apparent permeability (P (app)) values, as a measure of intestinal absorption, to the structures of two different data sets of drug candidates. Several molecular descriptors of the compounds were calculated and the optimal subsets were selected using a genetic algorithm; therefore, the method was indicated as Genetic Algorithm-Neural Network (GA-NN). A methodology combining a genetic algorithm search with neural network analysis applied to the modelling of Caco-2 P (app) has never been presented before, although the two procedures have been already employed separately. Moreover, we provide new Caco-2 cell permeability measurements for more than two hundred compounds. Interestingly, the selected descriptors show to possess physico-chemical connotations which are in excellent accordance with the well known relevant molecular properties involved in the cellular membrane permeation phenomenon: hydrophilicity, hydrogen bonding propensity, hydrophobicity and molecular size. The predictive ability of the models, although rather good for a preliminary study, is somewhat affected by the poor precision of the experimental Caco-2 measurements. Finally, the generalization ability of one model was checked on an external test set not derived from the data sets used to build the models. The result obtained is of interesting practical application and underlines that the successful model construction is strictly dependent on the structural space representation of the data set used for model development.

Application of data mining approaches to drug delivery

Computational approaches play a key role in all areas of the pharmaceutical industry from data mining, experimental and clinical data capture to pharmacoeconomics and adverse events monitoring. They will likely continue to be indispensable assets along with a growing library of software applications. This is primarily due to the increasingly massive amount of biology, chemistry and clinical data, which is now entering the public domain mainly as a result of NIH and commercially funded projects. We are therefore in need of new methods for mining this mountain of data in order to enable new hypothesis generation. The computational approaches include, but are not limited to, database compilation, quantitative structure activity relationships (QSAR), pharmacophores, network visualization models, decision trees, machine learning algorithms and multidimensional data visualization software that could be used to improve drug delivery after mining public and/or proprietary data. We will discuss some areas of unmet needs in the area of data mining for drug delivery that can be addressed with new software tools or databases of relevance to future pharmaceutical projects.

Chemical substituent effect on pyridine permeability and mechanistic insight from computational molecular descriptors

The objective was (1) to evaluate the chemical substituent effect on Caco-2 permeability, using a congeneric series of pyridines, and (2) compare molecular descriptors from a computational chemistry approach against molecular descriptors from the Hansch approach for their abilities to explain the chemical substituent effect on pyridine permeability. The passive permeability of parent pyridine and 14 monosubstituted pyridines were measured across Caco-2 monolayers. Computational chemistry analysis was used to obtain the following molecular descriptions: solvation free energies, solvent accessible surface area, polar surface area, and cavitation energy. Results indicate that the parent pyridine was highly permeable and that chemical substitution was able to reduce pyridine permeability almost 20-fold. The substituent effect on permeability provided the following rank order: 3-COO- < 4-NH2 < 3-CONH2 < 3-Cl < 3-CHO < 3-OH < 3-CH2OH < 3-C6H5 < 3-NH2 < 3-CH2C6H5 < 3-C2H5 < 3-H < 3-CH3 < 3-F < 4-C6H5. This substituent effect was better explained via molecule descriptors from the computational chemistry approach than explained by classic descriptors from Hansch. Computational descriptors indicate that aqueous desolvation, but not membrane partitioning per se, dictated substituent effect on permeability.

Prediction of ADMET Properties

This Review describes some of the approaches and techniques used today to derive in silico models for the prediction of ADMET properties. The article also discusses some of the fundamental requirements for deriving statistically sound and predictive ADMET relationships as well as some of the pitfalls and problems encountered during these investigations. It is the intension of the authors to make the reader aware of some of the challenges involved in deriving useful in silico ADMET models for drug development.

Computational prediction of oral drug absorption based on absorption rate constants in humans

Models for predicting oral drug absorption kinetics were developed by correlating absorption rate constants in humans (K(a)) with computational molecular descriptors. The K(a) values of a set of 22 passively absorbed drugs were derived from human plasma time-concentration profiles using a deconvolution approach. The K(a) values correlated well with experimental values of fraction of dose absorbed in humans (FA), better than the values of human jejunal permeability (P(eff)) which have previously been used to assess the in vivo absorption kinetics of drugs. The relationships between the K(a) values of the 22 structurally diverse drugs and computational molecular descriptors were established with PLS analysis. The analysis showed that the most important parameters describing log K(a) were polar surface area (PSA), number of hydrogen bond donors (HBD), and log D at a physiologically relevant pH. Combining log D at pH 6.0 with PSA or HBD resulted in models with Q(2) and R(2) values ranging from 0.74 to 0.76. An external data set of 169 compounds demonstrated that the models were able to predict K(a) values that correlated well with experimental FA values. Thus, it was shown that, using a combination of only two computational molecular descriptors, it is possible to predict with good accuracy the K(a) value for a new drug candidate.

Computational approaches for modeling human intestinal absorption and permeability

Human intestinal absorption (HIA) is an important roadblock in the formulation of new drug substances. Computational models are needed for the rapid estimation of this property. The measurements are determined via in vivo experiments or in vitro permeability studies. We present several computational models that are able to predict the absorption of drugs by the human intestine and the permeability through human Caco-2 cells. The training and prediction sets were derived from literature sources and carefully examined to eliminate compounds that are actively transported. We compare our results to models derived by other methods and find that the statistical quality is similar. We believe that models derived from both sources of experimental data would provide greater consistency in predictions. The performance of several QSPR models that we investigated to predict outside the training set for either experimental property clearly indicates that caution should be exercised while applying any of the models for quantitative predictions. However, we are able to show that the qualitative predictions can be obtained with close to a 70% success rate.

Prospective exploration of synthetically feasible, medicinally relevant chemical space

We describe a novel approach to direct the exploration of chemical space in an effort to balance synthetic accessibility and medicinal relevancy prior to experimental work. Reaction transforms containing empirical reactivity and compatibility information are dynamically assembled into reaction sequences (vProtocols) utilizing commercially available starting material feedstock. These vProtocols are evolved and optimized by a genetic algorithm, which leverages fitness functions based on predicted properties of generated molecular products. We present the underlying concepts, methodology and initial results of this prospective approach.

4SCan/vADME: intelligent library screening as a shortcut from hits to lead compounds

Managing to solve the first step in drug discovery - the hit finding - can be a quite elaborate task, but it is only the initial step to the final goal; hit-to-lead optimisation still lies ahead and consumes even more time and resources. The solution is rather simple, that is, to take only the most promising compounds into account; but who is going to decide which ones are the most promising among a list of tens of millions of compounds in a virtual combinatorial library? 4SCan/vADME helps by bridging the gap between virtual (combinatorial) libraries designed by chemists and the in silico methods, docking and alignment, for screening databases. After choosing a random starting set, the implemented learning and prediction algorithm iteratively considers only combinations of fragments that have shown to result in more suitable interactions by the chosen method. ADME properties of the final list are then calculated via several in silico methods, resulting in a combined evaluation of the individual compound's target-specific, as well as ADME, properties. Based on the latter list of evaluated compounds, medicinal chemists can then decide which compounds might be the best ones to synthesise first and to serve as possible lead candidates. Following a brief introduction to virtual high-throughput screening techniques, the 4SCan/vADME method is outlined and discussed in this paper, using an example coming out of the 4SC pipeline.

Prediction of drug absorption based on immobilized artificial membrane (IAM) chromatography separation and calculated molecular descriptors

The aim of this study was to evaluate the usefulness of IAM chromatography in building a model that would allow prediction of drug absorption in humans. The human intestinal absorption values (%HIA) for 52 drugs with low to high intestinal absorption were collected from the literature. The retention (capacity factor, k') of each drug was measured by reverse-phase HPLC using an IAM.PC.DD2 column (prepared with phosphatidylcholine analogs, 12 microM, 300A, 15 cm x 4.6 mm) with an eluent of acetonitrile-0.1M phosphate buffer at pH 5.4. In addition, 76 molecular descriptors and solubility parameters for each drug were calculated using ChemSW from the 3D-molecular structures. Stepwise regression was employed to develop a regression equation that would correlate %HIA with molecular descriptors and k'. Human intestinal absorption was reciprocally correlated to the negative value of the capacity factor (-1/k') (R=0.64). The correlation was further improved with the addition of molecular descriptors representing molecular size and shape (molecular width, length and depth) solubility (solubility parameter, HLB, hydrophilic surface area) and polarity (dipole, polar surface area) (R=0.83). Experimentally measured IAM chromatography retention values and calculated molecular descriptors and solubility parameters can be used to predict intestinal absorption of drugs in humans. Developed QSAR can be used as a screening method in the designing of drugs with appropriate IA and for the selection of drug candidates in the early stage of drug discovery process.

Ensemble of Linear Models for Predicting Drug Properties

We propose a new classification method for the prediction of drug properties, called random feature subset boosting for linear discriminant analysis (LDA). The main novelty of this method is the ability to overcome the problems with constructing ensembles of linear discriminant models based on generalized eigenvectors of covariance matrices. Such linear models are popular in building classification-based structure-activity relationships. The introduction of ensembles of LDA models allows for an analysis of more complex problems than by using single LDA, for example, those involving multiple mechanisms of action. Using four data sets, we show experimentally that the method is competitive with other recently studied chemoinformatic methods, including support vector machines and models based on decision trees. We present an easy scheme for interpreting the model despite its apparent sophistication. We also outline theoretical evidence as to why, contrary to the conventional AdaBoost ensemble algorithm, this method is able to increase the accuracy of LDA models.