David C. Whitley

Unsupervised forward selection: a method for eliminating redundant variables

An unsupervised learning method is proposed for variable selection and its performance assessed using three typical QSAR data sets. The aims of this procedure are to generate a subset of descriptors from any given data set in which the resultant variables are relevant, redundancy is eliminated, and multicollinearity is reduced. Continuum regression, an algorithm encompassing ordinary least squares regression, regression on principal components, and partial least squares regression, was used to construct models from the selected variables. The variable selection routine is shown to produce simple, robust, and easily interpreted models for the chosen data sets.

Use of Automatic Relevance Determination in QSAR Studies Using Bayesian Neural Networks

We describe the use of Bayesian regularized artificial neural networks (BRANNs) coupled with automatic relevance determination (ARD) in the development of quantitative structure-activity relationship (QSAR) models. These BRANN-ARD networks have the potential to solve a number of problems which arise in QSAR modeling such as the following: choice of model; robustness of model; choice of validation set; size of validation effort; and optimization of network architecture. The ARD method ensures that irrelevant or highly correlated indices used in the modeling are neglected as well as showing which are the most important variables in modeling the activity data. The application of the methods to QSAR of compounds active at the benzodiazepine and muscarinic receptors as well as some toxicological data of the effect of substituted benzenes on Tetetrahymena pyriformis is illustrated.

Biological data mining with neural networks: implementation and application of a flexible decision tree extraction algorithm to genomic problem domains

In the past, neural networks have been viewed as classification and regression systems whose internal representations were extremely difficult to interpret. It is now becoming apparent that algorithms can be designed which extract understandable representations from trained neural networks, enabling them to be used for data mining, i.e. the discovery and explanation of previously unknown relationships present in data. This paper reviews existing algorithms for extracting comprehensible representations from neural networks and describes research to generalize and extend the capabilities of one of these algorithms. The algorithm has been generalized for application to bioinformatics datasets, including the prediction of splice site junctions in Human DNA sequences. Results generated on this datasets are compared with those generated by a conventional data mining technique (C5) and conclusions drawn.

Van der Waals surface graphs and molecular shape

A van der Waals surface graph is the graph defined on a van der Waals surface by the intersections of the atomic van der Waals spheres. A van der Waals shape graph has a vertex for each atom with a visible face on the van der Waals surface, and edges between vertices representing atoms with adjacent faces on the van der Waals surface. These are discrete invariants of three‐dimensional molecular shape. Some basic properties of van der Waals surface graphs are studied, including their relationship with the Voronoi diagram of the atom centres, and a class of molecular embeddings is identified for which the dual of the van der Waals surface graph coincides with the van der Waals shape graph.

Sharpening the Toolbox of Computational Chemistry: A New Approximation of Critical F-Values for Multiple Linear Regression

Multiple linear regression is a major tool in computational chemistry. Although it has been used for more than 30 years, it has only recently been noted within the cheminformatics community that the standard F-values used to assess the significance of the resulting models are inappropriate in situations where the variables included in a model are chosen from a large pool of descriptors, due to an effect known in the statistical literature as selection bias. We have used Monte Carlo simulations to estimate the critical F-values for many combinations of sample size (n), model size (p), and descriptor pool size (k), using stepwise regression, one of the methods most commonly used to derive linear models from large sets of molecular descriptors. The values of n, p, and k represent cases appropriate to contemporary cheminformatics data sets. A formula for general n, p, and k values has been developed from the numerical estimates that approximates the critical stepwise F-values at 90%, 95%, and 99% significance levels. This approximation reproduces both the original simulated values and an interpolation test set (within the range of the training values) with an R2 value greater than 0.995. For an extrapolation test set of cases outside the range of the training set, the approximation produced an R2 above 0.93.

Quantitative structure-property relationships for predicting sorption of pharmaceuticals to sewage sludge during waste water treatment processes

Understanding the sorption of pharmaceuticals to sewage sludge during waste water treatment processes is important for understanding their environmental fate and in risk assessments. The degree of sorption is defined by the sludge/water partition coefficient (Kd). Experimental Kd values (n = 297) for active pharmaceutical ingredients (n = 148) in primary and activated sludge were collected from literature. The compounds were classified by their charge at pH 7.4 (44 uncharged, 60 positively and 28 negatively charged, and 16 zwitterions). Univariate models relating log Kd to log Kow for each charge class showed weak correlations (maximum R2 = 0.51 for positively charged) with no overall correlation for the combined dataset (R2 = 0.04). Weaker correlations were found when relating log Kd to log Dow. Three sets of molecular descriptors (Molecular Operating Environment, VolSurf and ParaSurf) encoding a range of physico-chemical properties were used to derive multivariate models using stepwise regression, partial least squares and Bayesian artificial neural networks (ANN). The best predictive performance was obtained with ANN, with R2 = 0.62–0.69 for these descriptors using the complete dataset. Use of more complex Vsurf and ParaSurf descriptors showed little improvement over Molecular Operating Environment descriptors. The most influential descriptors in the ANN models, identified by automatic relevance determination, highlighted the importance of hydrophobicity, charge and molecular shape effects in these sorbate-sorbent interactions. The heterogeneous nature of the different sewage sludges used to measure Kd limited the predictability of sorption from physico-chemical properties of the pharmaceuticals alone. Standardization of test materials for the measurement of Kd would improve comparability of data from different studies, in the long-term leading to better quality environmental risk assessments.

A solid-phase extraction method for rapidly determining the adsorption coefficient of pharmaceuticals in sewage sludge

The partitioning of pharmaceuticals in the environment can be assessed by measuring their adsorption coefficients (Kd) between aqueous and solid phases. Measuring this coefficient in sewage sludge gives an indication of their partitioning behaviour in a wastewater treatment plant and hence contributes to an understanding of their subsequent fate. The regulatory approved method for measuring Kd in sewage sludge is the US Environmental Protection Agencys Office of Prevention, Pesticides and Toxic Substances (OPPTS) guideline 835.1110, which is labour intensive and time consuming. We describe an alternative method for measuring the Kd of pharmaceuticals in sewage sludge using a modified solid-phase extraction (SPE) technique. SPE cartridges were packed at different sludge/PTFE ratios (0.4, 6.0, 24.0 and 40.0% w/w sludge) and eluted with phosphate buffer at pH 7.4. The approach was tested initially using three pharmaceuticals (clofibric acid, diclofenac and oxytetracycline) that covered a range of Kd values. Subsequently, the sorption behaviour of ten further pharmaceuticals with varying physico-chemical properties was evaluated. Results from the SPE method were comparable to those of the OPPTS test, with a correlation coefficient of 0.93 between the two approaches. SPE cartridges packed with sludge and PTFE were stable for up to one year; use within one month reduced variability in measurements (to a maximum of 0.6 log units). The SPE method is low-cost, easy to use and enables the rapid measurement of Kd values for a large number of chemicals. It can be used as an alternative to the more laborious full OPPTS test in environmental fate studies and risk assessments.

Conformation and dynamics of human urotensin II and urotensin related peptide in aqueous solution

Conformation and dynamics of the vasoconstrictive peptides human urotensin II (UII) and urotensin related peptide (URP) have been investigated by both unrestrained and enhanced-sampling molecular-dynamics (MD) simulations and NMR spectroscopy. These peptides are natural ligands of the G-protein coupled urotensin II receptor (UTR) and have been linked to mammalian pathophysiology. UII and URP cannot be characterized by a single structure but exist as an equilibrium of two main classes of ring conformations, open and folded, with rapidly interchanging subtypes. The open states are characterized by turns of various types centered at K8Y9 or F6W7 predominantly with no or only sparsely populated transannular hydrogen bonds. The folded conformations show multiple turns stabilized by highly populated transannular hydrogen bonds comprising centers F6W7K8 or W7K8Y9. Some of these conformations have not been characterized previously. The equilibrium populations that are experimentally difficult to access were estimated by replica-exchange MD simulations and validated by comparison of experimental NMR data with chemical shifts calculated with density-functional theory. UII exhibits approximately 72% open:28% folded conformations in aqueous solution. URP shows very similar ring conformations as UII but differs in an open:folded equilibrium shifted further toward open conformations (86:14) possibly arising from the absence of folded N-terminal tail-ring interaction. The results suggest that the different biological effects of UII and URP are not caused by differences in ring conformations but rather by different interactions with UTR.

Effect of sewage sludge type on the partitioning behaviour of pharmaceuticals: a meta-analysis

Assessment of the fate of pharmaceutical residues in the environment involves the measurement or prediction of their sewage sludge partition coefficient (Kd). Sewage sludge can be classified into four types: primary, activated, secondary and digested, each one with different physical and chemical properties. Published studies have measured Kd for pharmaceuticals in a variety of sludge types. This paper discusses the variability of reported Kd values of pharmaceuticals in different types of sewage sludge, using a dataset generated from the literature. Using a meta-analysis approach, it was shown that the measured Kd values depend on the type of sludge used in the test. Recommendations are given for the type of sludge to be used when studying the partitioning behaviour of pharmaceuticals in waste water treatment plants. Activated sludge is preferred due to its more homogenous nature and the ease of collection of consistent samples at a plant. Weak statistical relationships were found between Kd values for activated and secondary sludge, and for activated and digested sludge. Pooling of Kd values for these sludge types is not recommended for preliminary fate and risk assessments. In contrast, statistical analyses found stronger similarities between Kd values reported for the same pharmaceutical in primary and activated sludges. This allows the pooling of experimental values for these two sludge types to obtain a larger dataset for modelling purposes.

Vicinity analysis: a methodology for the identification of similar protein active sites

Vicinity analysis (VA) is a new methodology developed to identify similarities between protein binding sites based on their three-dimensional structure and the chemical similarity of matching residues. The major objective is to enable searching of the Protein Data Bank (PDB) for similar sub-pockets, especially in proteins from different structural and biochemical series. Inspection of the ligands bound in these pockets should allow ligand functionality to be identified, thus suggesting novel monomers for use in library synthesis. VA has been developed initially using the ATP binding site in kinases, an important class of protein targets involved in cell signalling and growth regulation. This paper defines the VA procedure and describes matches to the phosphate binding sub-pocket of cyclin-dependent protein kinase 2 that were found by searching a small test database that has also been used to parameterise the methodology.