Chemistry Reference
In-Depth Information
molecule, this term was called
cyclicity
. The four databases ACD,
[
35
]
NCI,
[
14
]
CMC
[
29
]
and
MDDR
[
36
]
were analyzed according to this scaffold-based classification approach. Only the
orally available drugs of CMC and MDDR were used. Adiversity map was constructed that
mapped complexity values against cyclicity values for each compound. Libraries that are
more diverse have a wider spread on this map. An interesting outcome was the ranking of
the four libraries according to chemical diversity. The ACD was most diverse, followed by
the NCI, then the CMC and finally the MDDR. Two factors contribute to the low diversity
of the MDDR: the majority of compounds are analogues and all compounds comply with
the 'drug-likeness'property values. Molecules contributing to the high diversity of theACD
included RNAs/DNAs and fullerene C
60
. Another interesting finding was that the orally
active drugs from the CMC and MDDR were distributed in a narrower region than the other
libraries.
8.3.3 Biological Activity
Sheridan
[
22
]
used common substructures to find fragment replacements in (drug-like)
molecules. For this, 98 445 drug-like molecules from the MDL Drug Data Report
(MDDR)
[
36
]
database were clustered according to similar biological activity, resulting in
556 clusters. Compounds from the same cluster were compared to find the 'highest-scoring
common substructure' (HSCS).
[
21
]
Only compounds with an HSCS significantly larger than
two randomly selected molecules of the same size were used to extract the fragments pairs
that differed. Two different methods were used to extract replacement fragment pairs. The
first method used atom-wise comparison of fragments, i.e. based on element and hybridiz-
ation of atoms. The second method also considered possible rings that the atoms were in
and adjacent functional groups, such as -NO
2
, -CO, -SO
2
or -PO
3
. Many of the classical
replacements in medicinal chemistry were found.
[
22
]
With atom type, substitution of C with
N in an aromatic ring (e.g. phenyl versus pyridine) was the most common. The next most
common was replacement of -O- with -S- in both rings and chains, followed by -N- with
-O- in rings, chains and esters versus amides.Another interesting commonly found replace-
ment was the change between a five- a six-membered ring. Also considering the context
of atoms in the comparison, e.g. a ring or functional group yielded a qualitatively similar
fragment list. For a more complete list of replacements, the reader is referred to Sheridan.
[
22
]
In a subsequent study, Sheridan
[
23
]
utilized the HSCS to identify fragments that are
associated with multiple biological activities. Sheridan considered activity in the widest
sense, ranging from
in vivo
biological effects (e.g. anti-hypertensive) to
in vitro
measures
(e.g. affinity for a receptor). Since high specificity is very much desired for new drugs,
knowledge about multi-activity fragments may be useful to avoid chemical classes likely
to have unwanted side-effects. On the other hand, scaffolds that are active on a variety of
receptors may form an attractive starting point in combinatorial library design. Pairs of
molecules with similar structure and dissimilar activity were identified first. For each pair,
the highest scoring common substructure (HSCS) was derived.
[
21
]
Again, only those HSCSs
were kept that were significantly larger than would be expected for two randomly selected
molecules. A 'consensus substructure' was generated from each molecule and its HSCS. It
consists of atoms that are considered to be 'conserved', i.e. atoms that appeared relatively
often in the set of HSCSs for that molecule. The most interesting consensus substructures
are those that are found in many molecules and have many unique activities. Therefore, the
generated consensus substructures were ranked according to both frequency of occurrence
