Biomedical Engineering Reference
In-Depth Information
reasons mentioned above (and others discussed in the following), the analysis of dif-
ferent landscape views is attractive for medicinal chemistry applications. However,
the choice of molecular representations and the assessment of compound similar-
ity, which are important factors for activity landscape quality and interpretability
(as discussed further below), present significant challenges for computational mod-
eling. The generation of activity landscapes is a topic of research at the interface
of chemoinformatics and medicinal chemistry, and so is the large-scale analysis of
SARs, which is highly relevant for medicinal chemistry but would not be possible
without specialized computational methods. Accordingly, the following discussion is
divided into three sections. In the first section, numerical SAR analysis functions are
introduced that depart from classical QSAR formalisms [4], are developed specif-
ically to capture quantitative SAR information contained in large compound data
sets and are also of central relevance for activity landscape modeling. In the second
section, alternative activity landscape representations are discussed that are based
on different design philosophies and have different information content. In the third
section, the discussion focuses on activity cliffs, and potential applications of the
activity cliff concept are critically evaluated.
16.2 NUMERICAL SAR ANALYSIS FUNCTIONS
Different types of SAR analysis function have been introduced [5]. However, these
numerical functions generally have in common that they systematically compare
chemical similarity and potency of specifically active compounds in a pairwise man-
ner. It is important to note that the assessment of compound similarity plays a role here
that differs fromother applications of similarity analysis in the chemoinformatics field
[6]. In many of these applications, molecular similarity is quantified as a measure of
activity similarity, although no well-defined quantitative relationship exists between
calculated similarity values and biological activity [6]. However, molecular similarity
calculated as a putative measure of activity similarity follows the similarity property
principle stating that “similar compounds should have similar bioactivity” [7]. This
principle is a central paradigm in chemoinformatics. For numerical SAR analysis
functions, similarity evaluation plays a different role because these functions also
explicitly involve the comparison of potency values. Thus, for this purpose, utilizing
reliable and robust representations and measures of structural similarity/relatedness
(see below) is much more important than finding representations that are particularly
relevant for a given bioactivity. In fact, consistent use of chosen molecular repre-
sentations and similarity measures is an essential condition for the comparability of
quantitative SAR assessment across different compound classes.
16.2.1 Structural Similarity vs. Activity Similarity
A prototypic SAR analysis function provided the basis for the generation of structure-
activity similarity (SAS) maps [8], which represent the first activity landscape rep-
resentation as defined herein, (discussed) further below. Structural similarity and
