1. Prof. Thanh N. Truong
University of Utah
Astonis LLC
Institute of Computational Science and Technology
2. The Drug Discovery Process
Drug Target
Identification
Target
Validation
Lead
Identification
Lead
Optimization
Pre-clinical &
Clinical
Development
It takes about 15 years and costs around 880 millions
USD, ~10,000 compounds (DiMasi et al. 2003;
Dickson & Gagnon 2004) to develop a new drug.
FDA Review
4. Genomics Facts
Around 99% of our genes have counterparts in mice
Our genetic overlap with chimpanzees is about 97.5%
The genetic difference between one person and another is less than 0.1 %
But because only a few regions of DNA actively encode life functions, the
real difference between one person and another is only 0.0003 %
It is becoming increasingly evident that the complexity of
biological systems lies at the level of the proteins, and that
genomics alone will not suffice to understand these systems.
5. Structure-based Computer-Aided Drug Design
Drug Target
Identification
Target
Validation
Shorten development
time to Lead
Identification
Lead
Identification
Lead
Optimization
Known 3D structure
PDB databank
Pre-clinical &
Clinical
Development
FDA Review
Unknown 3D structure for the target protein
Homology modeling or
Protein Structure Prediction
Reduce cost
MD simulations
Past Successes
1.
2.
3.
HIV protease inhibitor
amprenavir (Agenerase)
from Vertex & GSK (Kim et
al. 1995)
HIV: nelfinavir (Viracept) by
Pfizer (& Agouron) (Greer et
al. 1994)
Influenza neuraminidase
inhibitor zanamivir
(Relenza) by GSK
(Schindler 2000)
Target Model
Docking
Simulation
Cluster
Analysis
Scoring
Analysis
Trajectories
7. Docking
The Problem:
Determine the optimal binding structure of a ligand (a drug
candidate, a small molecule) to a receptor (a drug target, a protein or
DNA) and quantify the strength of the ligand-receptor interaction.
1.
2.
3.
4.
5.
Where the ligand will bind?
How will it bind?
How strong?
Why?
What make a ligand binds
to the receptor better than
the others?
6. ????
8. The Challenge
Ligand and receptor are
conformational flexible.
Receptor may have more than
one possible binding site.
Weak short-range Interactions:
hydrogen bonds, salt bridges,
hydrophobic contacts,
electrostatics, van der Walls
repulsions
Surface
complementary.
Binding affinity is the difference to
the uncomplexed state – solvation
and desolvation play important
role.
Binding affinity describes an
ensemble of complexes not a
single one.
Orientation
of Ligand
Bound waters
Flexibility of residues in
the binding site
Large protein
conformation change
9. Binding Affinity
Association equilibrium constant
[ RL ]s
Ka =
[ R ]s [ L ]s
b
∆Gg
+
∆Gsolv ( R )
∆Gsolv ( L )
Free energy of binding:
b
∆Gs = ∆H − T ∆S = − RT ln K a
Enthalpy Entropy
+
From the thermodynamic cycle:
b
b
∆Gs = ∆Gg + ∆Gsolv ( RL ) − {∆Gsolv ( R ) + ∆Gsolv ( L )}
b
∆Gs
∆Gsolv ( RL )
10. Docking Process
Descriptions of the
receptor 3D structure,
binding site and ligand
Sampling of the
configuration space of
the binding complex
Multiple binding
configurations for a
single protein
structure and a ligand
Evaluating free energy
of binding for scoring
Local/global minimum
Ensemble of protein
structures and/or
multiple ligands
11. Description of Receptor 3D Structure
Known 3D protein structures from Protein Data Bank (PDB)
(http://www.rcsb.org/pdb)
Locations of hydrogen atoms, bound water molecules, and metal ions are either not
known or highly uncertain.
Identities and locations of some heavy atoms (e.g., ~1/6 of N/O of Asn & Gln, and N/C
of His incorrectly assigned in PDB; up to 0.5 Å uncertainty in position)
Conformational flexibility of proteins is not known
Homology models from highly similar sequences with known structures
Critical analysis of the receptor structure before docking is needed:
resolution, missing residues, bound waters and ions, protonation states, etc.
12. Descriptions of Binding Site
Known binding site – PDB
database has about 6000 proteinligand complexes
Atomistic based
o Receptor atomic coordinates and
location of a binding box
Descriptor based
o
o
o
o
Surface
Volume
Points & distances, bond vectors
grid and various properties such
as electrostatic potential,
hydrophobic moment, polar,
nonpolar, atom types, etc
Unknown binding site
Blind docking with the binding box
cover the entire receptor –
computationally expensive
Better method for finding potential
binding sites is needed
13. Ligand Chemical Space
National Cancer Institute (NCI) public database
(http://129.43.27.140/ncidb2/)
About 250 K 3D structures
ZINC public database (http://zinc.docking.org/)
About 8 million 3D structures
PubChem public database (http://pubchem.ncbi.nlm.nih.gov/)
About 19 million entries (but no 3D structures)
Cambridge Structure database (CSD)
About 3 million crystal structures
Chemical Abstract Service (CAS) and SciFinder
Several other smaller databases …
Atomic partial charges
from MM force fields
or MO calculations
must be added to
each molecule for
evaluation of the
score function
14. Different Approaches in Docking
Complete conformation and configuration space are too large. Different approaches were
developed for effective sampling of the receptor-ligand configuration space.
Automated
Manual
Descriptor Matching
Simulation-based
• Use pattern-recognizing
geometric methods to
match ligand and receptor
site descriptors
• Ligand flexibility is limited
• Receptor is rigid
• Accuracy is not very good
– not discriminative
• Fast
• Use simulation methods to
sample the local configuration
space: MC-Simulated
Annealing, Genetic Algorithm.
Must run an ensemble of
starting orientations for
accurate statistics
• Ligand and protein flexibility
can be considered
• Free energy of binding is
evaluated
• Accuracy is good
• Time consuming
• Grid map is often used to
speed up energy evaluations
User interactive force
feedbacks through
haptic devices
Focus
15. MC-Simulated Annealing Method
Randomly change the receptor flexible residues,
ligand position, orientation, and/or conformation
Evaluate the new energy, Enew
YES
Enew < Eold ?
NO
Accept the new move
with P = exp{-∆E/kbT}
Accept the new move
Enew
Eold
Reduce the temperature
NO
Naccept or reject > Nlimit
YES
Done
16. Genetic Algorithm
Darwin Theory of Evolution
Living organisms
Made up of cells
Has the same set of chromosomes (DNA)
Genome: A set of all chromosomes
Chromosome consists of genes
Genotype: A particular set of genes
Each gene encodes a protein (a trait)
Each gene has a location in the chromosome (locus)
Reproduction by cross-over and mutation
17. Genetic Algorithm for Docking
Gene 1
Gene 2
Gene 3
x1 y1 z1 φ1 ψ1 ω1 τ1 τ2 τ3 τ4
Position
Orientation
Chromosome 1
Torsional angles
x2 y2 z2 φ2 ψ2 ω2 τ1’ τ2’ τ3’ τ4’ Chromosome 2
A chromosome is a possible solution: binding position,
orientation, and values of all rotatable torsional angles
Fitness Test
Translates genotypes to phenotypes (receptor-ligand complex
structures) for binding free energy evaluation.
A cell is a set of
possible solutions,
i.e. chromosomes.
Typical population
= 100-200
Select best parents
Those with large negative ∆G binding
Generate new generation
Migration: Move the best genes to the next generation
Cross-over: Exchange a set of genes from one parent chromosome to another. Typical cross-over
rate = 80-90%
Mutation: Randomly change a value of a gene, i.e. position, orientation, or torsional values. Typical
mutation rate = 0.5-1%
19. Lamarkian Genetic Algorithm -- AutoDock
Environmental adaptation of an individual’s phenotypic characteristics
acquired during lifetime can become heritable traits
Survival of the fittest.
1. Mutation and cross-breeding to
generate new genotype
generate
new possible ligand binding
configuration
2. Transfer to phenotype to evaluate
fitness
forming receptor-ligand
configuration.
Environmental
adaption
3. Adapt to the local environment to
improve fitness local minimization.
4. Transfer back to genotype for future
generations save the optimized
ligand binding configuration for future
generations.
Genetic Algorithm – Local Search
Morris et al., J. Comp. Chem. 1998, 19, 1639
Transfer to genotype
for future generations,
i.e. heritable traits.
GA
LS
20. Scoring Functions
Force Field based function
Focus
GOAL: Fast & Accurate
Experimentally observed complex
•Score = -∆Gbinding
•Has physical basis
•Fast with pre-computed grid
Multivariate regression fit physically
motivated structural functions to
experimentally known complexes
with measured binding affinity
-Score
Empirical function
Knowledge-based function
Statistical pair potential derived
from known complex structures
Descriptor based function
Based on chemical properties,
pharmacaphore, contact, shape
complementary
Complex configurations
21. Force Field Based Scoring Function
b
Score = −∆Gs
b
∆Gs = Cvdw ∗ ∆Gvdw + Cele ∗ ∆Gele + Chb ∗ ∆Ghb + Ctor ∗ ∆Gtor + Csolv ∗ ∆Gsolv
Coefficients are empirically determined using linear regression analysis from a set of
protein-ligand complexes from LPDB with known experimental binding constants.
23. Docking with Science Community Laboratory
Identify a target
Millions of molecules
from ZINC database
Docking simulation
with AutoDock-Vina
Rank according to
binding energy