Descriptors types

Linear representations are by far the most frequently used descriptor type. Apart from the already mentioned structural keys and hashed fingerprints, other types of information are stored. For example, the topological distance between pharmacophoric points can be stored [179, 180], auto- and cross-correlation vectors over 2-D or 3-D information can be created [185, 186], or so-called BCUT [187] values can be extracted from an eigenvalue analysis of the molecular adjacency matrix. 

Thus, it is still uncommon to test QSAR models (characterized by a reasonably high q ) for their ability to predict accurately biological activities of compounds not included in the training set. In contrast to such expectations, it has been shown that if a test set with known values of biological activities is available for prediction, there exists no correlation between the LOO cross-validated and the correlation coefficient between the predicted and observed activities for the test set (Figure 16.1). In our experience [17, 28], this phenomenon is characteristic of many datasets and is independent of the descriptor types and optimization techniques used to develop training set models. In a recent review, we emphasized the importance of external validation in developing reliable models [18]. 

Commonly used molecular descriptor types are listed. For each category, one or two representative examples are given. Dimensionality refers to the molecular representation (molecular formula, 2D drawing, or 3D conformation) from which the descriptors are calculated (adapted from ref. 4). 

The table shows a number of representative descriptor types (there are many more) that can be used to define chemical spaces. Each descriptor adds a dimension (with discrete or continuous value ranges) to the chemical space representation (e.g., selection of 18 descriptors defines an 18-dimensional space). Axes of chemical space are orthogonal only if the applied molecular descriptors are uncorrelated (which is, in practice, hardly ever the case). 

Many thousands of descriptor types have been developed over the years of chemical research. This short review is not intended to summarize these efforts in detail. Todeschini and Consonni have already provided a reference book for this purpose. Commercial and academic software for generating molecular descriptors has proliferated over the recent years. Todeschini s book lists software packages known in the year 2000. Speed of calculation, descriptor quality and diversity should be considered in selecting software. For example, absorption, distribution, metabolism, excretion, toxicity (ADMET) Predictor (Simulations Plus, Inc.) works at a rate of 250,000 molecules per horn calculating 272 molecular and 44 atomic descriptors in the following categories ... 

Descriptor type Molecular descriptors Number of descriptors... 

Figure 7 Localisation of most important descriptor types in terms of substituent position (ortho meta and para)...

Recently, this fragment constant method has been implemented in a routine and interfaced to ADAPT (45,46). Thus, the estimated log P value for a molecule can be used as a descriptor along with all the other descriptor types. 

These later two models of bioavailability as a continuous variable are linear since they used stepwise multiple linear regression (M LR) as the modeling tool. An obvious alternative, which may offer improved performance, is a nonlinear technique and such a model using an artificial neural network (ANN) was reported by Turner and colleagues [30], This study employed 167 compounds characterized by several descriptor types, ID, 2D, and 3D, and resulted in a 10-term model. Although the predictive performance was judged adequate, it was felt that the model was better able to differentiate qualitatively between poorly and highly bioavailable compounds. 

Let us evaluate the different properties and applications of a molecular descriptor while keeping the aforementioned requirements for descriptors in mind. We will focus on a particular descriptor type the radial distribution function (RDF). RDF descriptors grew out of the research area of structure-spectrum correlations but are far more than simple alternative representations of molecules. The flexibility of these functions from a mathematical point of view allows applying them in several other contexts. This chapter will give a theoretical overview of RDF descriptors as well as their application for the characterization of molecules, in particular for similarity and diversity tasks. 

RDF descriptors may be used in any combination to fit the required task. For instance, it is possible to calculate a multidimensional descriptor based on bond-path distances and restricted to nonhydrogen atoms in the shape of a frequency pattern. Consequently, more than 1,400 different descriptors are available. A final summary of RDF descriptor types, their properties, and applications is given in Table 5.1. This section summarizes typical applications, some of which are described in detail in the next chapter. 

Due to the characteristic shape, almost any raw RDF descriptor is skewed It typically exhibits an asymmetric tailing and is leptokurtic (flatted in relation to the Ganssian distribntion knrtosis > 0). Depending on the descriptor type, the size, and the symmetry of a molecule, the skewness or kurtosis of a raw RDF descriptor may also show asymmetric fronting or platykurtic behavior (peaked in relation to the Ganssian distribntion kurtosis < 0). As the general behavior applies to most of the RDF descriptors, it is no fault to neglect this skewness and to assume a skewed standard distribntion within the descriptor set. 

Earlier work in this area of shape analysis focused on QSAR studies accounting for conformational features of molecules, such as interatomic distances [89], explicit atomic coordinate sets [90], computed intermolecular distances [91], and simpler shape descriptors such as molecular volume [92]. Each of these descriptor types formally requires conformational analysis, and therefore produces, accordingly, a family of solutions for most structures. 

So, the only firm conclusions that can be drawn from these various investigations of different descriptor types is that combinations can be good and that the choice of descriptor, including the selection of subsets and possible combinations, is a vital part of the overall model building process. 

Different ways exist to classify molecular descriptors. We will adopt the classification of Downs.Most common descriptor types are largely atom based. These descriptors are based on the individual atoms composing the molecule, but with their description extended to incorporate information about the environment of the atom. To introduce the radically different path taken in quantum similarity as opposed to the more classic procedures in most similarity studies, a small presentation of different types of molecular descriptors first will be given. 

Mol2D FDA Freely available software that calculates more than 700 descriptors types based on a two-dimensional structure of molecules. http //www.fda.gov/ScienceReseareh/Bio-informaticsTools/Mold2/default.htm... 

The risk strategy is also based on selecting the method of superior accuracy out of four methods according to the resnlts of a leave-one-out cross-validation of the training set, but it is performed for each one of 11 levels of QL representation separately, with consideration to the descriptor type. This strategy implements a model of supremum consensus. There is no final voting procedure because each one of 44 prediction sets is treated as an independent information space. 

Geometrical descriptors are derived from the three-dimensional representations and include the principal moments of inertia, molecular volume, solvent-accessible surface area, and cross-sectional areas. Since conformational analysis (see Conformational Analysis 1 Conformational Analysis 2 and Conformational Analysis 3) often requires calculation of atomic charges, these routines can also produce electronic descriptors. Electronic descriptors characterize the molecular structures with such quantities as LUMO and HOMO energies, bond orders, partial atoim c charges, etc. Hybrid descriptors combine aspects of several of these descriptor types. The design and implementation of new descriptors is one important aspect of on-going research in the area of QSPR. 

---