Transcript Hierarchical Database Screenings for HIV

Hierarchical Database Screenings for HIV-1
Reverse Transcriptase Using a Pharmacophore
Model, Rigid Docking, Solvation Docking, and
MM-PB/SA
Junmei Wang, Xinshan Kang, Irwin D.Kuntz, and Peter A. Kollman
Encysive Pharmaceuticals Inc.
University of California, San Francisco
Presentation by Susan Tang
CS 379A
Background
•
There are two approaches to identifying drug leads
–
De novo design
• Aimed to design novel compounds that have electrostatic and hydrophobic
properties complementary to target
• Requires 3D structures of drug targets
–
Database screening
• Applies filters to identify potential drug leads from databases
• Can be divided into query-based and scoring-function-based methods
• Only scoring-function-based methods requires 3D structures of drug targets
1) Query-based screenings
- Search queries such as MW, #H-bond donors/acceptors, and pharmacophore
models are applied to database
- Computationally efficient since 3D structures are not used
- Wrong query fields may produce too high/too low # of hits
2) Scoring-function based approaches
- Apply target functions (typically free-energy calculations of inhibitor binding
to target) to obtain hits
- The most rigorous and accurate methods of free energy calculation are FEP
(Free energy perturbation) and TI (Thermodynamic integration)  but they
are too computationally intensive and thus not appropriate for DB screening
- There are several alternative methods as well (such as MM-PB/SA)
Purpose
•
Purpose: To develop a method for the identification of HIV-1 RT drug leads
using hierarchical database screening
•
Sequential Methods Used
1) Pharmacophore model
2) Multiple-conformation rigid docking
3) Solvation docking
4) MM-PB/SA (Molecular Mechanics-Poisson-Boltzman/surface area)
•
Significance of HIV-1 Reverse Transcriptase
– Important target in AIDS-related drug design
– Biological role is to transcribe viral RNA into dsDNA, which is necessary for
viral replication
– Recently, many crystal structures of NNRTI’s (non-nucleoside reverse
transcriptase inhibitors) with HIV-1 RT have been solved
– Since 3-D structures are available, HIV-1 RT poses as a good target for
drug lead development/screening
By showing that their methodology is accurate for HIV-1 RT, the authors
hope to demonstrate that the method can be widely applied to other
systems where target 3D structures are available.
Method Outline and Evaluation
Database = Refined ACD (Available Chemical Directory) DB of 150,000 compounds
Evaluation Criteria for Database Screening Performance
1) Hit rate =
known inhibitors that passed filter(s)
total number of known inhibitors in database
2) Enrichment factor = (Hit rate) x total number of compounds in database
total number of hits that passed filter(s)
Computational Methods
Filter 1: Pharmacophore Model
What is a pharmacophore model?
Defined as the three-dimensional arrangement of atoms - or groups of atoms –
responsible for the biological activity of a drug molecule.
19 crystal structures of HIV-1 RT
in complex with NNRTI’s
tri-feature pharmacophore model
Computational Methods
Filter 1: Pharmacophore Model
head
wing
wing
tail
•
•
•
19 HIV-1RT/NNRTI crystal structures were superimposed on PDB structure
1uwb (HIV-1 RT/TBO)
Spheres indicate where inhibitor atoms reside
Overall shape of bound inhibitors is like a butterfly (allosteric binding site of
enzyme)
Computational Methods
Filter 1: Pharmacophore Model
•
•
•
•
•
Tri-featured pharmacophore model designed from the “butterfly” shape
X1 : represents a 5 or 6 membered aromatic ring
X2 : represents a 5 to 7 membered ring
X3 : represents nitrogen, oxygen, or sulfur
Distinct distance patterns were also identified
Computational Results
Filter 1: Pharmacophore Model
-
Average RMSD of the 19 superimposed NNRTI’s = 0.86 angs.
40,000 compounds / 150,000 passed this filter
Hit rate = 95 %
Enrichment factor = 3.56
Computational Methods
Filter 2: Multiple-Conformation Rigid Docking
•
•
•
•
•
•
Spheres, where inhibitor atoms could potentially be, were highlighted on
HIV-1 RT/TBO reference structure
Cluster analysis selected one cluster consisting of 30-40 spheres around the
binding site and chose this as a center for docking
Conformational searches for the hits having passed Filter 1
Average Number of searched conformations for each molecule = 30
Rigid Docking was performed for all conformations
Crucial docking parameters:
1) Maximum orientations = 1000
2) Minimum matching nodes = 4
3) Maximum matching nodes = 15
4) No intramolecular score
5) Dielectric constant = 4.0
Computational Results
Filter 2: Multiple Conformation Rigid Docking
-
Average RMSD of the 19 superimposed NNRTI’s = 0.86 angs.
16,000 compounds / 40,000 had atleast 1 conformation that passed this filter
Hit rate = 76 %
Enrichment factor = 1.89
Computational Methods
Filter 3: Solvation Docking
•
•
•
•
•
•
•
Solvation docking parameters in the binding free energy formula could easily
vary from system to system
To derive solvation docking model specific for HIV-1 RT, a training set of 12
known HIV-1 RT/NNR-TI crystal structures were used
Each molecule in training set had an RMSD < 3.0 angstroms between the
docked and crystal structure
Parameters (alpha, beta, gamma) in formula I were optimized to reproduce
experimental binding free energies
Formula I:
Solvation docking was performed for molecules having passed filter II using a
solvation docking program
Program outputs the following terms:
1)VDW energy (hydrophobic interaction)
2)Screened electrostatic energy
3)Polar and non-polar accessible surface areas
Using derived solvation docking model, binding free energies were calculated
Computational Results
Filter 3: Solvation Docking
-
The solvation docking model with the following coefficients was produced
( Alpha = 0.1736, beta = 0.1709, gamma = 0.0049 )
-
Solvation docking model achieved average unsigned and rms errors of 1.03
and 1.16 kcal/mol between deltaG(calc) and deltaG(expt) for the training set
Computational Results
Filter 3: Solvation Docking
-
3360 compounds / 16,000 passed this filter with a threshold of –8.8 kcal/mol
Hit rate = 79 %
Enrichment factor = 3.74
Computational Methods
Filter 4: MM-PB/SA
•
•
First 3 filters: only ligand flexibility was taken into account
Current filter: application of MD simulations to sample conformational space
of BOTH inhibitor and receptor
•
•
For each molecule, MD simulations were done at 300 K with 2.0 fs time step
MD simulations carried out using this formula:
•
The inhibitor, water molecules, and receptor residues that are within 20
angs. Of inhibitor mass center were allowed to move during the simulations
Equilibration for 50 ps  20 snapshots were collected
For each snapshot: MM-PB/SA analysis was performed to calculate binding
free energy
•
•
Computational Results
Filter 4: MM-PB/SA
-
-
Because this is the most time/resource demanding step, MM-PB/SA was only
done on the 22 molecules in the control set & 30 top hits that passed Filter 3
16 / 22 control hits from Filter 3 yielded MM-PB/SA scorese < - 6.8 kcal/mol
10 / 30 top hits tested yielded MM-PB/SA scores < - 6.8 kcal/mol
Best hit had a binding free energy of – 17.7 kcal/mol (likely to be a real HIV1 RT inhibitor)
Summary
Results
•
Overall, 16/37 known NNRTs survived all filters
•
Overall hit rate = 41 %
•
Hit rate (first 3 filters) = 56 %
•
Enrichment rate (first 3 filters) = 25
Translates to: the probability of finding a real inhibitor randomly
from the hits of the first 3 filters is 25 fold higher than from the
whole database
Conclusion
The hierarchical multiple-filter database searching strategy attained both
high efficiency and high reliability, making it a viable option for drug lead
discovery.
Future Development
•
Making the time/resource limiting step, MM-PB/SA, more efficient
1) Run MD simulations using implicit (rather than explicit) water models
such as GB/SA and PB/SA
2) Development of new algorithm to calculate entropy accurately and
efficiently