Machine-learning techniques

This chapter gives a general introduction into the data analysis methodology. 

In recent decades, computer scientists have tried to provide computers with the ability to learn. This area of research was summarized under the umbrella term machine learning . Today machine learning is defined as the study of computer algorithms that improve automatically through experience [1]. 

The area of machine learning is thus quite broad, and different people have different notions about the domain of machine learning and what kind of techniques belong to this field. We will meet a similar problem of defining an area and the techniques involved in the field of data mining , as discussed in Section 9.8. We will use the term machine learning in this chapter to collect aU the methods that involve learning from data. 

One application of machine learning is that a system uses sample data to build a model which can then be used to analyze subsequent data. Learning from exam- 

---

Gelemter and Rose [25] used machine learning techniques Chapter IX, Section 1.1 of the Handbook) to analyze the reaction center. Based on the functionalities attached to the reaction center, the method of conceptual clustering derived the features a reaction needed to possess for it to be assigned to a certain reaction type. A drawback of this approach was that it only used topological features, the functional groups at the reaction center, and its immediate environment, and did not consider the physicochemical effects which are so important for determining a reaction mechanism and thus a reaction type. 

To understand the recommendations for structure descriptors in order to be able to apply them in QSAR or drug design in conjunction with statistical methods or machine learning techniques. 

Figure 9-1 shows the disciplines that contribute to machine learning techniques. 

The following sections present a more detailed description of the methods mentioned above. An overview of machine learning techniques in chemistry is given in Chapter IX, Section 1 in the Handbook. 

Several nonlinear QSAR methods have been proposed in recent years. Most of these methods are based on either ANN or machine learning techniques. Both back-propagation (BP-ANN) and counterpropagation (CP-ANN) neural networks [33] were used in these studies. Because 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 in which the activity surface was generated from the activities of training set compounds based on some predefined mathematical functions [34]. 

Inductive learning by decision trees is a popular machine learning technique, particularly for solving classification problems, and was developed by Quinlan (1986). A decision tree depicting the input/output mapping learned from the data in Table I is shown in Fig. 22. The input information consists of pressure, temperature, and color measurements of... 

The high dimensional nature of LIBS signals can lead to several computational issues when used in conjunction with many machine learning techniques. Dimensionality reduction is the process by which the high dimensional signals are mapped into a lower dimensional space. The resulting lower dimensional space can enable more robust performance when used in conjunction with pattern recognition techniques. 

Hert J, WiUett P, Wilton DJ, Addin P, Azzaoui K, Jacoby E, Schuffenhauer A. (2005) New Methods for Ligand-Based Virtual Screening Use of Data-Fusion and Machine-Learning Techniques to Enhance the Effectiveness of Similarity Searching. /. Chem. Inf. Model, (in the press). 

A more common use of informatics for data analysis is the development of (quantitative) structure-property relationships (QSPR) for the prediction of materials properties and thus ultimately the design of polymers. Quantitative structure-property relationships are multivariate statistical correlations between the property of a polymer and a number of variables, which are either physical properties themselves or descriptors, which hold information about a polymer in a more abstract way. The simplest QSPR models are usually linear regression-type models but complex neural networks and numerous other machine-learning techniques have also been used. 

Sakiyama, Y., Yuki, H., Moriya, T., Hattori, K., Suzuki, M., Shimada, K., Honma, T. Predicting human liver microsomal stability with machine learning techniques. J. Mol. Graph. Model. 2008,... 

Breneman, C., Bennett, Bi, J., Song, M., and Embrechts, M. (2002) New electron density-derived descriptors and machine learning techniques for computational ADME and molecular design. MidAtlantic Computational Chemistry Meeting, Princeton University, Princeton, NJ. 

One way to develop an in silica tool to predictive promiscuity is to apply a NB classifier for modeling, a technique that compares the frequencies of features between selective and promiscuous sets of compounds. Bayesian classification was applied in many studies and was recently compared to other machine-learning techniques [26, 27, 43, 51, 52]. 

Generally, tens of different machine-learning techniques are used in QSAR/QSPR studies. In this chapter, we describe only those that have been recently applied in studies of solvent extraction and complexation of metals. 

These results show that the new availability of high-quantity and -quality antibacterial peptide data allows more rigorous treatment and evaluation using machine learning techniques. In addition, the diversity of peptide sequences is dramatically improved—no longer is it necessary to start with a prototype... 

Beyond similarity-based applications, machine learning techniques may pick the specific descriptor elements that appear to correlate with the observed activity trends throughout a training set. Unlike in overlay models, where there is an obvious link between pharmacophore spheres or fields in space and their source atoms, the actual pairs (triplets, etc) of atoms in molecules that incarnate the picked descriptor elements must be first established, to gain any potential insights into the binding mechanisms. 

---