Ligand-Based Structural Hypotheses for Virtual Screening

1
Ligand-Based Structural Hypotheses for Virtual
Screening

• Ajay N. Jain
• Uses the tool described in the pervious paper

2
Agenda

• To investigate adequacy of the utility of a model
  comprised by the overlap of known ligands for a
  given target in identifying novel ligands with
  high sensitivity and specificity
• The targets structure is not known
• Justification Given a small number of
  potentially quite flexible molecules of diverse
  chemical structures, one must generate a
  hypothesis consisting of a single pose for each
  input molecule such that the joint superposition
  of all molecules will lead to predictive models
  of biological activity

3
Molecules Used
Chose 4 therapeutically interesting targets with
unknown three-dimensional structure, and
identified ligands known to associate with those
4
Positive Testing Set
5
Control Test Sets
HSV-1 Thymidine kinase inhibitors
Estrogen Receptor Ligands
6
Control Test Sets Alignments
7
GPCR Models
8
GABAA Model
9
ROC Curves
10
Examples of High Scoring Ligands
11
Selectivity of the Models
12
Serotonin Model Binding Affinity
13
Conclusions

• Offers a generally applicatble method for
  producing ligand-based binding site hypotheses,
  which can be used directly for high-throughput
  virtual screening or to form the basis on which
  to construct more detailed models of molecular
  activity
• Performance in terms of screening utility is
  comparable to that of many structure-based
  molecular docking techniques, but the best
  docking methods are capable of better sensitivity
  and specificity

14
Applicability

• where many existing ligands are known but where
  they share side-effects or biological properties
  that limit their biological utility
• where a small number of ligands have been
  discovered for a target that has not been
  extensively probed and augmentation of the set is
  a primary goal of a medicinal chemistry effort

15
Filler for Surflex Similarity Function