The fitness scores of the molecules with the dynamics-based models were also
found to be higher. To further confirm our observation, a docking-based virtual
screening was parallelly performed with the 41 CmaA1 snapshots and the reported
inhibitors. Docking with the MD CmaA1 snapshots not only could bind the most
active inhibitors as top scored hits, but also the docking scores were higher than the
ones with the crystal structure. These results thus throw light on the effect of
including multiple conformations of the targets on the screening abilities of the
pharmacophore models. Five out of the 40 dynamic e-pharmacophore models were
selected to be further used in our virtual screening study based on the consistency of
docking and pharmacophore screening results.
7.1.4 Dynamic Ligand-Based Pharmacophore Models:
Construction and Validation
Dynamic ligand-based pharmacophores were developed for the cofactors SAM and
SAHC considering their conformational heterogeneity in CmaA1 binding sites as
observed from MD trajectories of the respective model systems. Average structures
of SAM/SAHC were created after superimposing the conformations obtained from
each trajectory using uniform weighting method. Phase module of Schrodinger is
used to build the ligand-based pharmacophore models, each comprising six types
and 8–11 numbers of chemical features depending on the number and type of
interactions with the CmaA1 binding sites. To verify the screening efficiencies of
these models, a positive dataset of 23 CmaA1 inhibitors [96] and a negative dataset
of 1398 non-inhibitors (the same dataset used to validate the structure-based models
described in the previous section) were screened against each of the models. The
ligand-based models created using multiple conformations of the cofactors obtained
from the MD trajectories could screen up to 22 out of 23 CmaA1 active compounds
when the condition for matching was minimum four features of a model. The fitness
scores of the inhibitors matching the dynamic-ligand-based pharmacophore models
were also higher as compared to the one developed from the conformation of SAHC
bound to the crystal structure which was able to match to four CmaA1 inhibitors.
7.1.5 Pharmacophore-Based Virtual Screening
Once the best structure and ligand-based pharmacophore models were validated,
they were employed as filters in a novel virtual screening workflow consisting of
four different levels of screenings, viz. ligand-based pharmacophore mapping > structure-based pharmacophore mapping > docking > pharmacokinetic
properties (ADMET) filters. A focused library of 18,239 molecules from three
different sources was used for our virtual screening studies. As the first component
of the dataset, 6583 drugs reported in DrugBank were chosen, targeting drug
repurposing. The second component of the dataset was a set of 701 molecules
which were already reported to be highly active (<1 lM activity) on Mtb cells/
Pharmacophore Modelling and Screening: Concepts, Recent …
43
found to be higher. To further confirm our observation, a docking-based virtual
screening was parallelly performed with the 41 CmaA1 snapshots and the reported
inhibitors. Docking with the MD CmaA1 snapshots not only could bind the most
active inhibitors as top scored hits, but also the docking scores were higher than the
ones with the crystal structure. These results thus throw light on the effect of
including multiple conformations of the targets on the screening abilities of the
pharmacophore models. Five out of the 40 dynamic e-pharmacophore models were
selected to be further used in our virtual screening study based on the consistency of
docking and pharmacophore screening results.
7.1.4 Dynamic Ligand-Based Pharmacophore Models:
Construction and Validation
Dynamic ligand-based pharmacophores were developed for the cofactors SAM and
SAHC considering their conformational heterogeneity in CmaA1 binding sites as
observed from MD trajectories of the respective model systems. Average structures
of SAM/SAHC were created after superimposing the conformations obtained from
each trajectory using uniform weighting method. Phase module of Schrodinger is
used to build the ligand-based pharmacophore models, each comprising six types
and 8–11 numbers of chemical features depending on the number and type of
interactions with the CmaA1 binding sites. To verify the screening efficiencies of
these models, a positive dataset of 23 CmaA1 inhibitors [96] and a negative dataset
of 1398 non-inhibitors (the same dataset used to validate the structure-based models
described in the previous section) were screened against each of the models. The
ligand-based models created using multiple conformations of the cofactors obtained
from the MD trajectories could screen up to 22 out of 23 CmaA1 active compounds
when the condition for matching was minimum four features of a model. The fitness
scores of the inhibitors matching the dynamic-ligand-based pharmacophore models
were also higher as compared to the one developed from the conformation of SAHC
bound to the crystal structure which was able to match to four CmaA1 inhibitors.
7.1.5 Pharmacophore-Based Virtual Screening
Once the best structure and ligand-based pharmacophore models were validated,
they were employed as filters in a novel virtual screening workflow consisting of
four different levels of screenings, viz. ligand-based pharmacophore mapping > structure-based pharmacophore mapping > docking > pharmacokinetic
properties (ADMET) filters. A focused library of 18,239 molecules from three
different sources was used for our virtual screening studies. As the first component
of the dataset, 6583 drugs reported in DrugBank were chosen, targeting drug
repurposing. The second component of the dataset was a set of 701 molecules
which were already reported to be highly active (<1 lM activity) on Mtb cells/
Pharmacophore Modelling and Screening: Concepts, Recent …
43