Artificial neural networks in QSAR

Devillers J. A new strategy for using supervised artificial neural networks in QSAR. SAR QSAR Environ Res 2005 16 433 12. 

With the advent of powerful computers and easy access to them, and the introduction of expert systems, artificial intelligence, and neural networks in QSAR, radically different models designed from noncongeneric large sets of chemicals have been proposed. No attempts are made to design a model that is easily interpretable in terms of MOA. The main objective of the present models is to provide powerful simulators with a wide domain of application for predicting the toxicity of any kind of molecule. 

Polymeropoulos, E. E. (1993) QSAR analysis of time- and dose-dependent in vivo drag effects using artificial neural networks. Trends QSAR Mol. Modell. 92, Proc. Eur. Symp. Structure-Activity Relationships QSAR and Molecular Modeling 9th, pp. 546-549., Strasbourg, France. 

A challenging task in material science as well as in pharmaceutical research is to custom tailor a compound s properties. George S. Hammond stated that the most fundamental and lasting objective of synthesis is not production of new compounds, but production of properties (Norris Award Lecture, 1968). The molecular structure of an organic or inorganic compound determines its properties. Nevertheless, methods for the direct prediction of a compound s properties based on its molecular structure are usually not available (Figure 8-1). Therefore, the establishment of Quantitative Structure-Property Relationships (QSPRs) and Quantitative Structure-Activity Relationships (QSARs) uses an indirect approach in order to tackle this problem. In the first step, numerical descriptors encoding information about the molecular structure are calculated for a set of compounds. Secondly, statistical and artificial neural network models are used to predict the property or activity of interest based on these descriptors or a suitable subset. 

In the last decades not only thousands of chemical descriptors but also many advanced, powerful modeling algorithms have been made available, The older QSAR models were linear equations with one or a few parameters. Then, other tools have been introduced, such as artificial neural network, fuzzy logic, and data mining algorithms, making possible non linear models and automatic generation of mathematical solutions. 

Compared with the artificial neural network (ANN) approach used in previous work to predict CN12 the linear regression model by QSAR is as good or better and easier to implement. The predicted CN values, some of which are tabulated in Table 1, will be employed below to evaluate the different catalytic strategies to optimize the fuel. 

Fig. 2. Structure of an artificial neural network. The network consists of three layers the input layer, the hidden layer, and the output layer. The input nodes take the values of the normalized QSAR descriptors. Each node in the hidden layer takes the weighted sum of the input nodes (represented as lines) and transforms the sum into an output value. The output node takes the weighted sum of these hidden node values and transforms the sum into an output value between 0 and 1.

Stanton DT, Jurs PC (1990) Development and use of charged partial surface area structural descriptors in computer-assisted quantitative structure-property relationship studies. Anal Chem 62 2323—2329. Tetko iy Kovalishyn Vy Livingstone DJ (2001) Volume learning algorithm artificial neural networks for 3D-QSAR studies. J Med Chem 44 2411-2420. 

Clearly, the constant can be included into threshold value B, so that the function /o(C) = 1 is not necessary. We must stress that in such form the probabilistic approach has no tuned parameters at all. Some tuning of naive Bayes classifier can be performed by selection of the molecular structure descriptors [or /(C)] set. This is a wonderful feature in contrast to QSAR methods, especially to Artificial Neural Networks. 

An inexperienced user or sometimes even an avid practitioner of QSAR could be easily con-fiased by the multitude of methodologies and naming conventions used in QSAR studies. Two-dimensional (2D) and three-dimensional (3D) QSAR, variable selection and artificial neural network methods, comparative molecular field analysis (CoMFA), and binary QSAR present examples of various terms that may appear to describe totally independent approaches, which cannot be even compared to each other. In fact, any QSAR method can be generally defined as the application of mathematical and statistical methods to the problem of finding empirical relationships (QSARmod-els)of the form, D . D ), where... 

These considerations provide an impetus for the development of fast, nonlinear, variable selection QSAR methods that can avoid the aforementioned problems of linear QSAR. Several nonlinear QSAR methods have been proposed in recent years. Most of these methods are based on either artificial neural network (ANN) (50, 61, 137-142) or machine learning techniques (65,143-145). Given that optimization of many parameters is involved in these techniques, the speed of the analysis is relatively slow. More recently. Hirst reported a simple and fast nonlinear QSAR method (146), in which the activity surface was generated from the activities of training set compounds based on some predefined mathematical function. 

The computational construction of artificial neural networks has also been applied to relate physicochemical parameters of benzodiazepines with their receptor affinity and to predict BZR properties and BZR ligand affinities. In a study by Maddalena and Johnston, back-propagation artificial neural networks were used to examine the QSAR between substituent constants at six positions on 57 ben-zodiazepinones with their empirically determined binding affinities (118). Among the findings of the study were the following ... 

Methods other than thermodynamic cycles are often used to calculate acid dissociation constants. Previous publications implement the theoretical relationship between pKa and structural property [6], bond valence methods and bond lengths [33], pKa correlations with highest occupied molecular orbital (HOMO) energies and frontier molecular orbitals [34], and artificial neural networks [35] to predict pKa values. In addition much work has been done using physical properties as quantitative structure-activity relationship (QSAR) descriptors, and regression equations with such descriptors to yield accurate pKa values for specific classes of molecules [36-47]. The correlation of pKas to various molecular properties, however, is often restricted to specific classes of compounds, and it is... 

Hemmateenejad B, Akhond M, Miri R, Shamsipur M. Genetic algorithm applied to the selection of factors in principal component-artificial neural networks application to QSAR study of calcium channel antagonist activity of 1,4-dihydropyri-dines (nifedipine analogs). J Chem Inf Comput Sci 2003 43 1328-34. 

In terms of practical application, expert systems overlap with systems for deriving and applying quantitative structure-activity relationship (QSAR) models or equations, and with systems using artificial neural networks (ANN) or genetic algorithms. The expert systems described in this chapter are characterized by their use of a generalized store of knowledge. 

---