David W. Salt

Simultaneous prediction of aqueous solubility and octanol/water partition coefficient based on descriptors derived from molecular structure

It has been shown that water solubility and octanol/water partition coefficient for a large diverse set of compounds can be predicted simultaneously using molecular descriptors derived solely from a two dimensional representation of molecular structure. These properties have been modelled using multiple linear regression, artificial neural networks and a statistical method known as canonical correlation analysis. The neural networks give slightly better models both in terms of fitting and prediction presumably due to the fact that they include non-linear terms. The statistical methods, on the other hand, provide information concerning the explanation of variance and allow easy interrogation of the models. Models were fitted using a training set of 552 compounds, a validation set and test set each containing 68 molecules and two separate literature test sets for solubility and partition.

Regression analysis for QSAR using neural networks

Abstract Neural networks have been used to analyse QSAR data giving promising results. However, there is the danger of chance “correlations” and “over-fitting”. We have examined a reported analysis and shown that the size of the hidden layer can be reduced giving more efficient training while maintaining predictive performance.

Instance-specific multi-objective parameter tuning based on fuzzy logic

Finding good parameter values for meta-heuristics is known as the parameter setting problem. A new parameter tuning strategy, called IPTS, is proposed that is a novel instance-specific method to take the trade-off between solution quality and computational time into consideration. Two important steps in the method are an a priori statistical analysis to identify the factors that determine heuristic performance in both quality and time for a specific type of problem, and the transformation of these insights into a fuzzy inference system rule base which aims to return parameter values on the Pareto-front with respect to a decision maker’s preference.

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.

Variable selection and specification of robust QSAR models from multicollinear data: arylpiperazinyl derivatives with affinity and selectivity for α2-adrenoceptors

Two QSAR models have been identified that predict the affinity and selectivity of arylpiperazinyl derivatives for α1 and α2 adrenoceptors (ARs). The models have been specified and validated using 108 compounds whose structures and inhibition constants (Ki) are available in the literature [Barbaro et al., J. Med. Chem., 44 (2001) 2118; Betti et al., J. Med. Chem., 45 (2002) 3603; Barbaro et al., Bioorg. Med. Chem., 10 (2002) 361; Betti et al., J. Med. Chem., 46 (2003) 3555]. One hundred and forty-seven predictors have been calculated using the Cerius 2 software available from Accelrys. This set of variables exhibited redundancy and severe multicollinearity, which had to be identified and removed as appropriate in order to obtain robust regression models free of inflated errors for the β estimates – so-called bouncing βs. Those predictors that contained information relevant to the α2 response were identified on the basis of their pairwise linear correlations with affinity (−log Ki) for α2 adrenoceptors; the remaining variables were discarded. Subsequent variable selection made use of Factor Analysis (FA) and Unsupervised Variable Selection (UzFS). The data was divided into test and training sets using cluster analysis. These two sets were characterised by similar and consistent distributions of compounds in a high dimensional, but relevant predictor space. Multiple regression was then used to determine a subset of predictors from which to determine QSAR models for affinity to α2-ARs. Two multivariate procedures, Continuum Regression (the Portsmouth formulation) and Canonical Correlation Analysis (CCA), have been used to specify models for affinity and selectivity, respectively. Reasonable predictions were obtained using these in silico screening tools.

The kinetics of insecticide action. Part V: Deterministic models to simulate the movement of pesticide from discrete deposits and to predict optimum deposit characteristics on leaf surfaces for control of sedentary crop pests

Mathematical models have been used to identify combinations of deposit size, density and concentration which result in effective control of sedentary crop pests using minimal amounts of insecticide. A model based on point source diffusion gave an adequate description of the spread of biocidal areas around deposits with time ; the more complex disc source model gave similar results. The point source model has been developed further to investigate how pesticide inputs might be reduced while maintaining adequate control. Models based on the cumulative effects of toxicant with time gave marginally better fits in the tails of the tolerance distribution. Prediction of LN 50 values using a model which takes account of overlapping biocidal areas was in reasonable agreement with experimental results. Models which have been developed to investigate the factors which affect the control of sedentary crop pests by insecticides and acaricides may also be used to predict optimal spray patterns for contact herbicides and fungicides.

Neural networks in the search for similarity and structure — activity

This book is concerned with molecular similarity, what it is, how it may be quantified and how it can be used in the design of new drugs. Perhaps one of the most important aspects of similarity, however it is defined, is the question of how it may be perceived. One of the skills that we humans can lay particular claims to is our ability to recognize patterns, in other words to perceive similarity, and thus many of the similarity tools used in drug design are intended to express similarity, leaving the task of perception to the human ‘expert’. Unfortunately, our ability to recognize patterns is generally restricted to relatively low dimensional data sets (i.e. 2, 3 or 4D) and thus we need help when similarity is described by hundreds or thousands of variables. This help can take the form of a variety of statistical methods, some of the most useful being pattern recognition techniques (Livingstone, 1991a) which literally set out to identify any underlying patterns in sets of data. An alternative approach in trying to tackle the problem is by the use of comparative molecular field analysis (CoMFA) (see Chapter 12).

Pharmacokinetics: Computational Versus Experimental Approaches to Optimize Insecticidal Chemistry

An ideal insecticide is a compound that is highly toxic to target insect pest species but has a low toxicity to non-target organisms, including beneficial species of insects, and with a high differential toxicity between insects and vertebrates. It should also have a low environmental impact, and be rapidly removed from the environment when adequate control of pest species has been achieved. Very few compounds approach this ideal, and this gives an indication of the difficulty of designing insecticides to meet these diverse (often conflicting) criteria. Toxicity, defined as a capacity to cause injury in a living organism defined with reference to the quantity of substance administered or absorbed, the way in which the substance is administered and distributed in time (single or repeat doses), the type and severity of injury, the time needed to produce injury, the nature of the organism(s) affected, and other relevant conditions) (Nordberg et al. 2004), is easy to achieve, especially on in vitro screens. However, the combination of high specific whole organism toxicity and low toxicity to non-target organisms is much more difficult to achieve even within an identified chemical class. Currently, most of the commercially available insecticides act at a relatively small number of target sites, and most are nerve poisons causing disruption of nerve function by interfering with neurotransmitter action or by disrupting ion channel functioning. This limited number of target sites is associated with increasing problems of resistance, both through modified metabolic capabilities and modified target sites. There is an urgent need for insecticides with novel modes of action. However, the costs associated with the development of a new product are extremely high, and there is a need to provide useful information to aid the selection and modification of leads at an early stage. Some of the problems to be overcome are also encountered in the development of new pharmaceutical products. Useful biological activity at the site of action (adequate PD activity) is not necessarily reflected in high levels of activity when compounds are applied to whole organisms. In the case of insecticides, there is a further challenge not encountered in dosing subjects with therapeutic agents; the need to apply an appropriate dose on small animals, sometimes in flight, dispersed over a wide area such as a crop or an area of forest. In the case of both pharmaceuticals and insecticides, the

Joint Continuum Regression for Analysis of Multiple Responses

The rationale behind developing a multiple response algorithm for continuum regression (CR) is to provide the user with a method of investigating how any number of responses change simultaneously given one set of physico-chemical properties. The background behind multiple response algorithms is well documented with such algorithms available for ordinary least squares regression (OLS) often referred to as multivariate linear regression (MVLR), partial least squares regression (PLS), sometimes known as PLS2 and principal components regression (PCR). Furthermore, an algorithm has been proposed by Brooks and Stone [1994], named joint continuum regression (JCR), which maintains a number of the properties of their formulation of the single response continuum regression [Stone and Brooks, 1990].

Chemical Fingerprints Containing Biological and Other Nonstructural Data

The pharmaceutical industry is concerned with identifying drugs for safe treatment for human disease. Searching for these bio-active compounds is like looking for a needle in a haystack. The aim of this research is to develop a methodology for searching large databases for lead compounds using similarity, and chemical libraries for high throughput screening (HTS) using diversity. In this study, similarity has been calculated using the Tanimoto Index (TI) and diversity has been derived as (1-TI).