Chemistry Reference
In-Depth Information
computational methods were used to break down and analyse the constituent components
of a database of commercially available drug molecules. Molecules are split into ring
systems, linkers, side-chains and frameworks, where a framework is defined as the union
of ring systems and linkers in a molecule. A surprisingly small number of frameworks
(41), taking into consideration atom types and bond orders, describe the framework of
∼
24% of the molecules in the database. These frameworks, along with 30 of the most
common side-chains, were used in the process of selecting compounds from the ACD
(MDL Available Chemicals Directory) for inclusion in the SHAPES library. The final
library contained commercially available compounds that are water soluble at 1 mM, have
MW in the range 68-341 Da (average 194 Da), 6-22 heavy atoms and a
C
log
P
of -2.2
to 5.5. The library profile reflects the fact that the design was dominated by the selection
of suitable frameworks and side-chains and the requirement for high aqueous solubility.
More recent design strategies incorporate physicochemical property filters where it would
be unusual to pass compounds with such high log
P
values.
Breaking down drug-like molecules into fragments would at face value seem an obvious
starting point for a fragment library. One of the issues, however, associated with earlier
attempts to use fragments was that chemistry featured poorly in the fragmentation process,
leading to synthetic difficulties in subsequent application. Consequently, the breakdown of
drug-like compounds into fragments has been automated in computational methods, such
as RECAP,
[
21
]
that are chemically intuitive. Originally cited as a means of identifying priv-
ileged molecular building blocks for constructing combinatorial libraries rich in biological
motifs, these methods are equally suited to generating libraries of fragments for screening
and ensuing optimisation studies. Synthetic optimisability features heavily in Novartis's
design for a fragment library. First an analysis was undertaken looking at results fromNMR
screening as compared with HTS results in relation to the Hann complexity model.
[
22
]
The
analysis, which is discussed in more detail below, supports the basic principles of fragment-
based discovery
[
23
]
and provided the framework from which to design a next-generation
fragment library. Emphasis is placed on the ability to optimise low-affinity hits through
incorporating one or more synthetic handles. Investigation of previous attempts to utilize
synthetically tractable functionality in fragments highlighted the fact that in these smaller,
less complexmolecules the functional group is more often than not an integral component in
binding to the target protein. Strategies for increasing the likelihood that a synthetic handle is
available for modification include masking the functionality, selection of functionality that
is normally not recognised by a protein or simply incorporating multiple groups. Novartis
employed what they term a fragment pair strategy, where a fragment building block with
an exposed synthetic handle is transformed into a screening fragment by masking the func-
tional group but having minimal effect on a range of computed properties. The fragment
building block can then be used in subsequent elaboration when a hit is observed for the
corresponding fragment screening compound. In addition to library profiles akin to those
outlined above, Similog keys are used as a measure of complexity. Similog keys represent
pharmacophoric triplets and it was found that, as expected, fragments with micromolar to
millimolar potency are significantly lower in complexity than the more drug-like molecules
used in HTS.
Vernalis also use NMR screening in their strategy for fragment-based drug discovery
called SeeDs
[
24
]
(Selection of Experimentally Exploitable Drug start points). A pharma-
cophore fingerprint is used as a measure of chemical complexity and diversity amongst
