![[Feedback]](/icons/banner/feedbackACT.gif){kind=icon}
![[For Subscribers]](/icons/banner/subscriberACT.gif){kind=icon}
![[Archive]](/icons/banner/archiveACT.gif){kind=icon}
![[Search]](/icons/banner/searchACT.gif){kind=icon}
![[Contents]](/icons/banner/tocACT.gif){kind=icon}
|
|
|
|
|
||
| |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Department of Decision Sciences and Engineering Systems, Rensselaer Polytechnic Institute, Troy, New York (M.J.E.); and GeneGo, Inc., St. Joseph, Michigan, and Department of Pharmaceutical Sciences, University of Maryland, Baltimore, Maryland (S.E.)
(Received October 2, 2006; accepted November 29, 2006)
 |
Abstract |
|---|
 Top Abstract Materials and MethodsResults and DiscussionReferences |
|---|
), quantitative structure-activity relationships (QSARs) (Shen et al., 2003; Balakin et al., 2004), electronic models (Korzekwa et al., 2004), and commercial drug metabolism databases (Borodina et al., 2004), as well as other methods that have been comprehensively reviewed elsewhere (de Graaf et al., 2005; Ekins et al., 2005a; de Groot, 2006). Some approaches have combined metabolite data and rules for suggesting metabolic pathways across multiple species (Erhardt, 2003). Such databases may also be useful for calculating the probability for a given metabolic reaction (Boyer and Zamora, 2002) to then indicate potential metabolites and the sites of metabolism using statistical or algorithmic approaches (Borodina et al., 2004). Although these types of comprehensive databases generally enable numerous search options to retrieve molecule structures and published information, the predictive capabilities seem limited at present (Wishart et al., 2006). A major limitation is that they are unlikely to have a complete dataset of reactions and molecular structures to extrapolate for a new molecule. In turn, the user is reliant on the quality of the published in vitro or in vivo data which, in many cases, may predate modern analytical methods, such that older published metabolic pathways may be incomplete. In reality, such database approaches provide knowledge of most published data and are perhaps limited to interpolation.
The combination of different approaches to drug metabolite prediction may balance the strengths and weaknesses of each approach, and several commercial methods are now pursuing this direction. MetaDrug represents one such method, combining a manually annotated database of human drug metabolism information including xenobiotic reactions, enzyme substrates, and enzyme inhibitors with kinetic data (Ekins et al., 2005b, 2006). This database has enabled the generation of rules for predicting likely metabolic reactions. The parent molecule and metabolites may then be scored through integrated QSAR models and rules for molecule reactivity before visualizing molecules as nodes on a network diagram (Ekins et al., 2005b, 2006).
Such rule-based metabolite predictions indicate that it is possible to generate many more metabolites than have been identified in the literature, which may make the methods less useful (Ekins et al., 2006). We are therefore investigating approaches to limit the metabolites to those that are most likely. Recently, a number of machine learning approaches including support vector machines and kernel-partial least squares (K-PLS) (Rosipal and Trejo, 2001) have been implemented in a single software package (Analyze/StripMiner), and this package was used with several benchmark datasets (Bennett and Embrechts, 2003) including protein binding and other physicochemical properties. The results with K-PLS indicated that it could be favorably applied to other datasets to enable QSAR model construction and aid drug discovery research. In the current proof of concept study, we have used K-PLS to generate preliminary classification models to identify whether a metabolite is likely to be produced for a particular parent molecule.
|
Materials and Methods |
|---|
|
TopAbstractMaterials and MethodsResults and DiscussionReferences |
|---|
), and this represents a small fraction of the human drug metabolism content. These molecules were prepared as an sdf file containing data for the 65 metabolic pathways of interest (Ekins et al., 2005a) with binary data for the presence or absence of a metabolite.
Descriptor Calculation. ChemTree software (GoldenHelix, Bozeman, MT) running on a Pentium 4 processor was used to generate augmented atom molecular descriptors (Young et al., 2002) representing the presence or absence of a particular heavy atom with its immediately bonded neighbors. In total, 61 descriptors were generated for the set of molecules.
Data Preprocessing. Metabolic reactions with greater that two examples of the metabolite rule were then used for modeling; this narrowed down the dataset considerably. The matrix of molecular descriptors and biological activity data were then scaled (normalized) and variables with unchanging values were removed using feature selection with the Analyze/StripMiner software (software available from M.J.E. at http://www.rpi.edu/locker/82/001182/) (Embrechts et al., 2001). From the descriptors with more than 95% correlation between each other (i.e., "cousin descriptors"), only the descriptors most correlated with the response were retained. In addition, four sigma outliers were brought within 2.5 sigma.
K-PLS Modeling Method and Testing. The Analyze software uses the K-PLS method (Rosipal and Trejo, 2001) with two key parameters, the number of latent variables and the Parzen window or Gaussian kernel sigma. In this study, the number of latent variables is held fixed at 5, and the Gaussian kernel sigmas are tuned using a second-order Newton method in which the performance criterion is the error minimization on the validation data using 5-fold cross-validation. The sigmas were tuned just once, using the metabolite with the most positive instance cases. Sigma tuning on just one single metabolite is a conservative approach that prevents over-tuning. Furthermore, the fact that the model still has a good predictive power on the other metabolites is another indication that over-tuning did not occur in this case.
|
):
 |
). For the predictive modeling on the other metabolites, the same sigmas were used. Sigma-tuning also allows for an identification procedure for pointing out the most relevant attributes by considering that the attributes with the larger sigma values are less relevant. After sigma tuning, the individual metabolites were predicted using K-PLS with a Gaussian kernel with multiple sigmas, using a leave-one-out procedure. Because the number of positive examples of a metabolite generally was exceeded by the number of negative instances, the discrimination between positive and negative cases was made using a bias with a threshold of –0.5 for choosing the operating point on the receiver operator curve. The area under the curve (AUC) values were also calculated, with higher values approximating to better classifications. Because of the imbalance in the number of positive and negative examples, the balanced error rate was calculated taking the average of the number of correct that were positive and the number correct that were negative. In this case, higher numbers are preferable.
|
Results and Discussion |
|---|
|
TopAbstractMaterials and MethodsResults and DiscussionReferences |
|---|
). In silico metabolite prediction typically generates many more potential metabolites than are actually observed. The emergence of machine learning tools combined with databases of human metabolism information represent methods for producing more reliable predictions of metabolites from an input structure alone. In the current study, for each of the more than 300 molecules selected from the MetaDrug database with metabolism information, two-dimensional molecular descriptors were calculated. Twenty-three of the 65 reactions had sufficient binary data for modeling, and a K-PLS model was produced for each using the Analyze/StripMiner software (Bennett and Embrechts, 2003). We evaluated the resulting classification models for predicting metabolic reactions after leave-one-out testing (Table 1). In general, we found that the reactions that are well populated with literature data (e.g., N-dealkylation, aromatic and aliphatic hydroxylation, and O-glucuronidation) produced K-PLS models that perform well when assessed using the AUC value and the receiver operator curve plots (Table 1; Fig. 1). As expected, those models that are sparsely populated with few positive instances of a metabolite being observed corresponding to a particular reaction (generally non-cytochrome P450-related), are of poorer quality, indicating that no reliable classification can be made. Exceptions include N-hydroxylation and double bond peroxidation, in which there are remarkably few positive examples, but results are favorable for predictions, indicating that the examples provided generate useful rules based on path length descriptors. This preliminary work with both phase I and II reactions indicates that such an approach requires generally much larger databases than were used here, which will be available in later versions of MetaDrug. Despite this, K-PLS models for N-dealkylation, O-dealkylation, aromatic hydroxylation, aliphatic hydroxylation, O-glucuronidation, and O-sulfation reactions had AUC values between 0.75 and 0.84 and were able to predict between 61 and 79% active molecules upon leave-one-out testing while, more importantly, the balanced error predictions were between 70 and 82%. Therefore, this represents a useful method to classify the potential for an unknown molecule to undergo these particular metabolic reactions. However, this approach requires further testing using considerably more data for the many sparsely populated metabolic reactions. In addition, external validation of all models with large test sets of molecules will be required alongside measures to ensure that a prediction is reliable, such as those based on molecule similarity. This work represents the first occasion, to our knowledge, that K-PLS has been used for metabolite prediction, and the results obtained are promising with unbalanced datasets. The integration of this K-PLS approach with rule-based and other QSAR methods could result in a more effective method for metabolite prediction that would be useful in numerous drug discovery applications where reliable metabolite identification is important.
|
|
Footnotes |
|---|
Competing Financial Interest: MetaDrug is a proprietary tool developed and licensed by GeneGo, Inc.
Article, publication date, and citation information can be found at http://dmd.aspetjournals.org.
ABBREVIATIONS: QSAR, quantitative structure-activity relationship; K-PLS, kernel-partial least squares; AUC, area under the curve.
Address correspondence to: Dr. Sean Ekins, ACT LLC, 601 Runnymede Ave., Jenkintown, PA 19046. E-mail ekinssean{at}yahoo.com, sekins{at}arnoldllc.com
|
References |
|---|
|
TopAbstractMaterials and MethodsResults and DiscussionReferences |
|---|
This article has been cited by other articles:
|
|
 |
|
  K. Fenner, J. Gao, S. Kramer, L. Ellis, and L. Wackett Data-driven extraction of relative reasoning rules to limit combinatorial explosion in biodegradation pathway prediction Bioinformatics, September 15, 2008; 24(18): 2079 - 2085. [Abstract] [Full Text] [PDF]
|
|
| |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
 |
 |
 |
|
 |
 |
 |
