Quantative Structure-

Activity Relationships
 ?Why QSAR

• The number of compounds required for

synthesis in order to place 10 different groups
in 4 positions of benzene ring is 104
• Solution: synthesize a small number of
compounds and from their data derive rules to
predict the biological activity of other
compounds.
QSAR and Drug Design
Compounds + biological activity

QSAR

New compounds with

improved biological activity
 ?What is QSAR
A QSAR is a mathematical relationship
between a biological activity of a molecular
system and its geometric and chemical
characteristics.
QSAR attempts to find consistent relationship
between biological activity and molecular
properties, so that these “rules” can be used to
evaluate the activity of new compounds.
 Statistical Concepts

Input: n descriptors P1,..Pn and the value of

biological activity (EC50 for example) for
m compounds.
Bio P1 P2 ... Pn
Cpd1 0.7 3.7
Cpd2 3.2 0.4
…
Cpdm
 Statistical Concepts
• The problem of QSAR is to find coefficients
C0,C1,...Cn such that:
Biological activity = C0+(C1*P1)+...+(Cn*Pn)
and the prediction error is minimized for a list of
given m compounds.

• Partial least squares (PLS) is a technique used for

computation of the coefficients of structural
descriptors.
 3D-QSAR
Structural descriptors are of immense importance in
every QSAR model.
Common structural descriptors are pharmacophores and
molecular fields.
Superimposition of the molecules is necessary.
3D data has to be converted to 1D in order to use PLS.
 3D-QSAR Assumptions
The effect is produced by modeled compound and not it’s
metabolites.
The proposed conformation is the bioactive one.
The binding site is the same for all modeled compounds.
The biological activity is largely explained by enthalpic
processes.
Entropic terms are similar for all the compounds.
The system is considered to be at equilibrium, and kinetics
aspects are usually not considered.
Pharmacokinetics: solvent effects, diffusion, transport are
not included.
 QSAR and 3D-QSAR
Software
Tripos – CoMFA, VolSurf
MSI – Catalyst, Serius

Docking Software
DOCK – Kuntz
Flex – Lengauer
LigandFit – MSI Catalyst
 3D molecular fields

A molecular field may be represented by 3D

grid.
Each voxel represents attractive and repulsive
forces between an interacting partner and a
target molecule.
An interacting partner can be water, octanol or
other solvents.
Common 3D molecular fields
MEP – Molecular Electrostatic Potential (unit positive
charge probe).
MLP – Molecular Lipophilicity Potential (no probe
necessary).
GRID – total energy of interaction: the sum of steric
(Lennard-Jones), H-bonding and electrostatics (any
probe can be used).
CoMFA – standard: steric and electrostatic, additional:
H-bonding, indicator, parabolic and others.
Comparative Molecular Field
Analysis (CoMFA) - 1988

Compute molecular fields grid

Extract 3D descriptors
Compute coefficients of QSAR equation
 CoMFA molecular fields

A grid wit energy fields is calculated by placing

a probe atom at each voxel.
The molecular fields are:
Steric (Lennard-Jones) interactions
Electrostatic (Coulombic) interactions
• A probe is sp3 carbon atom with charge of +1.0
 CoMFA 3D-QSAR
Each grid voxel corresponds to two variables
in QSAR equation: steric and electrostatic.
The PLS technique is applied to compute the
coefficients.
Problems:
• Superposition: the molecules must be
optimally aligned.
• Flexibility of the molecules.
 3D-QSAR of CYP450cam
with CoMFA
• Training dataset from 15 complexes of
CYP450 with different compounds was used.
• The alignment of the compounds was done by
aligning of the CYP450
proteins from the
complexes.
3D-QSAR of CYP450cam
with CoMFA
Maps of electrostatic fields:
BLUE - positive charges
RED - negative charges
Maps of steric fields:
GREEN - space filling areas
for best Kd
YELLOW - space
conflicting areas
 VOLSURF

The VolSurf program predicts a variety of

ADME properties based on pre-calculated
models. The models included are:
• drug solubility
• Caco-2 cell absorption
• blood-brain barrier permeation
• distribution
 VOLSURF

• VolSurf reads or computes molecular fields,

translates them to simple molecular
descriptors by image processing techniques.
• These descriptors quantitatively characterize
size, shape, polarity, and hydrophobicity of
molecules, and the balance between them.
 VOLSURF Descriptors
• Size and shape: volume V, surface area S, ratio volume
surface V/S, globularity S/Sequiv (Sequiv is the surface area of a
sphere of volume V).
• Hydrophilic: hydrophilic surface area HS, capacity factor
HS/S.
• Hydrophobic: like hydrophilic LS, LS/S.
• Interaction energy moments: vectors pointing from the
center of the mass to the center of hydrophobic/hydrophilic
regions.
• Mixed: local interaction energy minima, energy minima
distances, hydrophilic-lipophilic balance HS/LS, amphiphilic
moments, packing parameters, H-bonding, polarisability.
 VOLSURF

hydrophobic (blue) and hydrophilic (red) surface

.area of diazepam
 Catalyst
Catalyst develops 3D models (pharmacophores) from a
collection of molecules possessing a range of diversity in
both structures and activities.
Catalyst specifies hypotheses in terms of chemical
features that are likely to be important for binding to the
active site.
Each feature consists of four parts:
Chemical function
Location and orientation in 3D space
Tolerance in location
Weight
 Catalyst Features
• HB Acceptor and Acceptor-Lipid
• HB Donor
• Hydrophobic
• Hydrophobic aliphatic
• Hydrophobic aromatic
• Positive charge/Pos. Ionizable
• Negative charge/Neg. Ionizable
• Ring Aromatic
 Catalyst HipHop
Feature-based pharmacophore modeling:
uses ONLY active ligands
no activity data required
identifies binding features for drug-receptor
interactions
generates alignment of active leads
the flexibility is achieved by using multiple
conformers
alignment can be used for 3D-QSAR analysis
 Catalyst HipoGen
Activity-based pharmacophore
modeling:
uses active + inactive ligands
activity data required (concentration)
identifies features common to actives
missed by inactives
used to “predict” or estimate activity
of new ligands
 Catalyst CYP3A4
substrates pharmacophore

Hydrophobic area, h-bond Saquinavir (most active compound)

donor, 2 h-bond acceptors fitted to pharmacophore
 Catalyst CYP2B6
substrates pharmacophore

hydrophobic areas, h-bond 3 ethoxy-4-trifluoromethylcoumarin-7

acceptor fitted to pharmacophore
 Catalyst Docking – Ligand Fit
Active site finding
Conformation search of ligand against site
Rapid shape filter
determines which
conformations should be scored
Grid-based scoring for those
conformations passing the filter
 Catalyst Docking – Ligand
Flexibility
Monte Carlo search in torsional space
Multiple torsion changes simultaneously
The random window size depends on the
number of rotating atoms
 Catalyst Docking – Scoring
pKi = – c – x (vdW_Exact/ Grid_Soft)
+ y (C+_pol)
– z (Totpol^ 2)

• vdW = softened Lennard-Jones 6-9 potential

• C+_pol = buried polar surface area involved in attractive
ligand-protein interactions
• Totpol^ 2 = buried polar surface area involved in both
attractive and repulsive protein-ligand interactions
 3D-QSAR of CYP450cam
with DOCK
Goal:
• Test the ability of DOCK to discriminate between
substrates and non-substrates.
Assumption:
• Non-substrate candidate is a compound that
doesn’t fit to the active site of CYP, but fits to the
site of it’s L244A mutant.
 Methods
Docking of 20,000 compounds to ‘bound’ structure of
CYP and L244A mutant.
11 substrate candidates were selected from 500 high
scoring compounds for CYP.
6 non-substrate candidates were selected from a
difference list of L244A and CYP.
Optimization of compounds 3D structures by SYBYL
molecular mechanics program and re-docking. As a result 2
compounds move from “non-substrate” list to “substrate”
list and one in the opposite direction.
 Prediction Results
All compounds predicted as “non-substrates” shown no
biological activity.
4 of the 11 molecules predicted as “substrates” were
found as non-substrates.
The predictions of DOCK are sensitive to the parameter
of minimum distance allowed between an atom of the
ligand and the receptor (penetration constrains).
Prediction Results
 References
• Cruciani et al., Molecular fields in quantitative structure-permeation relationships:
the VolSurf approach, J. Mol. Struct. (Theochem), 2000, 503:17-30
• Cramer et al.,Comparative Molecular Field Analysis (CoMFA). 1. Effect of shape
on Binding of steroids to Carrier proteins, J. Am. Chem. Soc. 1988, 110:5959-5967
• Ekins et al., Progress in predicting human ADME parameters in silico, J.
Pharmacological and Toxicological Methods 2000, 44:251-272
• De Voss et al., Substrate Docking Algorithms and Prediction of the Substrate
Specifity of Cytochrome P450cam and its L244A Mutant, J. Am. Chem. Soc. 1997,
119:5489-5498
• Ekins et al., Three-Dimensional Quantative Structure Activity Relationship
Analyses of Substrates for CYP2B6, J. Pharmacology and Experimental
Therapeutics, 1999, 288:21-29
• Ekins et al., Three-Dimensional Quantative Structure Activity Relationship
Analysis of Cytochrome P-450 3A4 Substrates, J. Pharmacology and Experimental
Therapeutics, 1999, 291:424-433
• Sechenykh et al., Indirect estimation of protein-ligand complexes Kd in database
searching, www.ibmh.msk.su/qsar/abstracts/sech.htm