Strategies for Compound Selection

The simplest and fastest techniques for grouping molecules are partitioning methods. Every molecule is represented by a point in an n-dimensional space, the axes of which are defined by the n components of the descriptor vector. The range of values for each component is then subdivided into a set of subranges (or bins). As a result, the entire multidimensional space is partitioned into a number of hypercubes (or cells) of fixed size and every molecule (represented as a point in this space) falls into one of the cells [20, 125-127]. 

Partition-based selection is much faster than clustering, mapping, or dissimilarity methods because no similarities of pairs of molecules have to be calculated. 

The computational complexity is only linearly proportional to the number of compounds that need to be processed. For this reason, the method can be used for rapid grouping of compounds in huge databases or libraries. The method becomes impractical if the number of dimensions in the descriptor set is too high (about seven dimensions or grouping parameters is the limit). 

The previous section introduced methods to group molecules based on the descriptor set Strategies for selecting compounds from these groups or clusters depend on the properties that are to be optimized. Objectives of compound selection are to  

Nonetheless, the quahty of a screening Hbrary will improve if it is expanded by designed Hbraries that are optimized for small redundancy and high internal diversity. A noticeable increase of hit rates can also be expected if the library design is integrated in the drug design cycle. 

Further discussion of other experimental design techniques, with the exception of simplex optimization (see next section), is outside the scope of this book. Hopefully this section will have introduced the principles of experimental design the reader interested in further details should consult one of the excellent texts available which deal with this subject in detail (see Box et al. 1978 Morgan 1991). A recent review discusses the application of experimental design techniques to chemical synthesis (Carlson and Nordahl 1993). 

Some of the earliest techniques for compound selection were essentially visual and as such have considerable appeal compared with the (apparently) more complex statistical and mathematical methods. The first method to be reported came from a study of the relationships between a set of commonly used substituent constants (Craig 1971). The stated purpose of this work was to examine the interdependence of these parameters and, as expected, correlations (see Box 2.1) were found between the hydrophobidty descriptor, %, and a number of bulk parameters such as molecular volume and parachor. Why should interdependence between substituent constants be important There are a number of answers to this question, as discussed further in this book, but for the present it is sufficient to say that interdependence between parameters is required so that clearer, perhaps mechanistic, conclusions might be drawn from correlations. As part of the investigation Craig plotted various parameters together, for example the plot of ct vs. n shown in Fig. 2.2 such plots have 

An important property of any variable, which is used in many statistical operations is a quantity called the variance, V. The variance is a measure of how the values of a variable are distributed about the mean and is defined by 

Division of the covariance by the square root of the product of the individual variances allows us to put a scale on the degree to which two variables are related. If y changes by exactly the same amount as jc changes, and in the same direction, the correlation coefficient is +1. If y decreases by exactly the same amount as x increases the correlation coefficient is —1. If the changes in y are completely unrelated to the changes in x, the correlation coefiScient will be 0. 

the two main problems in compound selection are the choice of analogues to sample effectively a multi-parameter space and the avoidance of collinearity between physicochemical descriptors. A number of methods have been proposed to deal with these two problems. An attractive approach was published by Hansch and co-workers (Hansch et al. 1973) which made use of cluster analysis (Chapter 5) to group 90 substituents described by five physicochemical parameters. Briefly, cluster analysis operates by the use of measurements of the distances between pairs of 

---

Olah, M. M., Bologa, C. G., Oprea, T. I. Strategies for compound selection. Curr. Drug Discov. Technol. 2004, 1, 211-220. 

Dunbar JB (2000) Compound acquisition strategies. Pac Symp Biocomput 5 552 562 Olah MM, Bologa CG, Oprea TI (2004) Strategies for compound selection. Curr Drug Discov Technol 1 211-220... 

The variation of the principal properties in a set of compounds is quantified by the score values. This variation can be displayed by plotting the scores of different components against each other. Such score plots are very useful for selecting test compounds for experimental studies. Strategies for the selection of test system based upon principal properties are discussed in Sect. 4.6. 

A useful strategy for the selective substitution of dianhydroglucitols utilizing a titanium-mediated hydride reduction of l,4 3,6-dianhydro-2,5-di-0-nitro-D-glucitol has appeared. Thus reduction of the latter with titanium(III) borohydride [Ti(BH4)3] affords the monoester 10, and with diisopropyloxytitanium(III) borohydride [( PrO)2TiBH4]the isomeric compound 11 is produced. ... 

The development of coherent strategies for the selective binding of analyte molecules, by rational design of synthetic receptors, remains one of chemistry s most challenging goals. Research conducted to this end is driven by a fundamental curiosity and the need to monitor compounds of industrial, environmental, and biological importance. 

---