Chemistry Reference
In-Depth Information
8
Computational Approaches to
Fragment and Substructure
Discovery and Evaluation
Eelke van der Horst and Adriaan P. IJzerman
8.1
Introduction
Nowadays, large molecular databases are easily accessible to the research community. This
is illustrated by the advent of free online resources such as PubChem
[
1
]
and eMolecules.
[
2
]
These publicly available databases consist of structure and property data for millions of
small molecules. Both databases are accessible through web-based search tools and are
therefore an unprecedented source of small-molecule data. Outside the public domain, sim-
ilar progress is taking place. Large molecular databases are becoming available that include
bioactivity data, for example WOMBAT
[
3
]
(WOrld of Molecular BioAcTivity). Molecular
data from these sources may be used to construct predictive models, such as structure-
activity/property relationships (SARs/SPRs) or classification models. These models can be
based on molecular properties, such as lipophilicity, solubility and molar weight, but also
on molecular structures
per se
.
In silico
fragmentation of molecular structures is often used
to provide a dataset of structural elements of the intact molecule. Analysis of the result-
ing fragments is useful to derive novel classifiers, e.g. for predicting the activity of new
molecules. What is meant by the term
fragment
depends on the context. In the chemical
sense, a fragment is a small, low molecular weight substance with weak affinity often used
to 'build'a higher affinity lead compound. This is different from the computational sense. In
the computational context, the term fragment or substructure denotes some structural part
