Introduction

Drugs
• Molecules that can be introduced
to change biological activity
Structure-Based
Drug Design

Thomas Funkhouser
Princeton University
CS597A, Fall 2005

Slide courtesy of Bill Welsh

Introduction Structure-Based Drug Design

Drug targets Goal:
• Enzyme - inhibitors • Given a protein structure,
• Receptor - agonists or antagonists and/or its binding site,
• Ion channels - blockers and/or its active ligand (possibly bound to protein),
• Transporter - update inhibitors find a new molecule that changes the protein’s activity
• DNA - blockers

Slide courtesy of Bill Welsh

Structure-Based Drug Design Structure-Based Drug Design

Receptor-based drug design: Ligand-based drug design:
• Given a protein structure, • Given an protein structure,
and/or its binding site, and/or its binding site,
and/or its active ligand (possibly bound to protein), and/or its active ligand (possibly bound to protein),
find a new molecule that changes the protein’s activity find a new molecule that changes the protein’s activity

HIV Protease Inhibitor

Example courtesy of Bill Welsh Example courtesy of Joe Corkery

1
Challenges Outline
Scoring of chemical models Virtual drug screening
• Activity • Ligand-based
• Toxicity § 2D structure matching
§ 3D structure matching
• “Druggability”
§ QSAR
• etc.
De novo drug design
Search of chemical space
• Models
• Add/remove/replace chemical groups § Simulation
• Conformations § Knowledge-based
• etc. • Construction algorithms
§ Incremental construction
§ Fragment-based
§ Stochastic optimization

Ligand-Based Drug Screening Ligand-Based Drug Screening

17b-Estradiol

HO
Androstenediol

HO
Progesterone

O
M oxestrol

HO
Coumestrol

HO O O
Strategies:
• 2D matching
O OH
OH OH
O
CH3
H 3C O OH
• 3D matching
Diethylstilbestrol Genistein M ethoxychlor
Bisphenol A
• Pharmacore matching
HO HO O H 3C O OC H 3 HO OH

• Quantitative Structure Activity Relationships (QSAR)

OH O
OH Cl Cl Cl

OH
Nafoxidine
a- Zearalanol(Zeranol) 4-Hydroxytamoxifen
O
HO O
N N
HO
OH

O H 3C O
OH
O

Tamoxifen
Clomifene ICI 164,384

O
N O HO (C H 2) 10C O N(C H3 )C 4H 9
N

Cl
OH

Ligands for Estrogen Receptor Slide courtesy of Bill Welsh

2D Structure Matching 2D Substructure Matching

O
O Query
N N
H O
N O O
H H H2N
N N OH H
N N NH2 N
N O
N Query N
N
O O
OH N N
H H2N
HO N
N O
H N
N N N N
N N O
N N N N N
N O N O
O
O

Slide courtesy of Bill Welsh Slide courtesy of Bill Welsh

2
2D Substructure Matching 2D Substructure Matching

N-N
O-C(-N)-C
CH3-Ar-CH3
C-N-N
N-Ar-Ar-O
N-C-O
N-Ar-O
OH > 1
CH3 > 1
N>1
NH
...

Which keys are present?

Slide courtesy of Bill Welsh Slide courtesy of Bill Welsh

2D Substructure Matching 2D Substructure Matching

*Activity data from Uehling et al, J. Med.Chem. 1995
Tanimoto coefficient:
2.5

(A∩B)
Activity difference (log units)

T=
( A ∪B)
2

1.5
struct A: 00010100010101000101010011110100 13 bits
struct B: 00000000100101001001000011100000 8 bits
1

A & B: 00000000000101000001000011100000 6 bits A ∩B

A or B: 00010100110101001101010011110100 15 bits A ∪B 0.5

T = 6 / 15 = 0.4 0
1.00 0.95 0.90 0.85

Tanimoto similarity
Slide courtesy of Bill Welsh Slide courtesy of Bill Welsh

Outline 3D Structure Matching

Virtual drug screening Search database of molecules for ones with
• Ligand-based similar 3D shape and chemistry
§ 2D structure matching
3D structure matching
§ QSAR
A B
De novo drug design
• Models
§ Simulation
§ Knowledge-based
• Construction algorithms
§ Incremental construction
§ Fragment-based
§ Stochastic optimization

3
3D Structure Matching 3D Substructure Matching
Search database of molecules for ones with O a = 8.62+- 0.58 Angstroms
N O
b = 7.08+
similar 3D shape and chemistry c a
- 0.56 Angstroms

c = 3.35+
- 0.65 Angstroms
O
O O

b N O O
O
S

N N
O O
Ai O O N
Bi O N
N
O
O O
N N N O
O O O P O O
N O N
N
O N O P O
O N O
O N
N O P O
O
O O O
2 2
d ( A, B) = Ai − B + A − Bi N O O
O O
Ai ∈A Bi ∈B
Slide courtesy of Bill Welsh

3D Substructure Matching 3D Substructure Matching

A
O(s1)
C(u)

O(s1)

3.2Å
6.7 3.3 - 4.3 Å Cl O H Cl
4.2-4.7 4.3Å
5.2 4.8 O
6.8 - 7.8 Å

5.1-7.1
O H
3.6 - 4.6 Å

Flexible
[O,S ]
A

Slide courtesy of Bill Welsh

Rigid Slide courtesy of Bill Welsh

3D Substructure Matching Pharmacore Matching

17b-Estradiol Androstenediol Progesterone M oxestrol Coumestrol

DISTANCE
DISTANCECONSTRAINTS
CONSTRAINTS HO HO O HO HO O O

(Qualitative
(QualitativeAffinity
Affinityprediction
predictionmostly)
mostly)
O OH
OH H 3C O OH
OH CH3
O

Bisphenol A
BIOPHASE Diethylstilbestrol Genistein M ethoxychlor

HO HO O H 3C O OC H 3 HO OH
A A'
OH O
OH Cl Cl Cl

OH
Nafoxidine
a- Zearalanol(Zeranol) 4-Hydroxytamoxifen

B B' RECEPTOR
RECEPTOR O
O
HO
N N
HO
OH

O H 3C O
OH
O

CC' Tamoxifen
Clomifene ICI 164,384

O
N O HO (C H 2) 10C O N(C H3 )C 4H 9
N

LIGAND
Cl
OH

Slide courtesy of Bill Welsh Ligands for Estrogen Receptor Slide courtesy of Bill Welsh

4
Pharmacore Matching Outline
17b-Estradiol

HO
Androstenediol

HO
Progesterone

O
M oxestrol

HO
Coumestrol

HO O O
Virtual drug screening
• Ligand-based
O
OH OH
O
CH3
H 3C O OH
OH
§ 2D structure matching
Diethylstilbestrol Genistein M ethoxychlor
Bisphenol A § 3D structure matching
OH QSAR
12 A
HO HO O H 3C O OC H 3 HO OH

OH O

OH
OH Cl Cl Cl
De novo drug design
a- Zearalanol(Zeranol) 4-Hydroxytamoxifen

O
Nafoxidine

O
• Models
HO
N N
HO
OH § Simulation
OH
O HO H 3C O
§ Knowledge-based
O

Tamoxifen
Clomifene ICI 164,384
• Construction algorithms
O
N O
N
HO (C H 2) 10C O N(C H3 )C 4H 9 § Incremental construction
§ Fragment-based
Cl
OH
§ Stochastic optimization
Ligands for Estrogen Receptor Slide courtesy of Bill Welsh

QSAR QSAR
Learn model for activity as function of descriptors Use model to predict activities for new leads
(properties) computed from molecules from their descriptors

Compound Activity (pKi) Descriptors (Xi)

"Y" Mol. Vol. (Å3) LogP Dipole Mom (µ
µ)
OH

1 2.34 420 2.8 0.97

2 1.89 332 4.6 2.23
3 0.23 198 -0.3 3.36
4 3.67 467 3.7 0.45 HO
5 2.55 359 -1.5 1.77 V logP µ
etc. etc. etc. etc. etc.
!
"#$
%& ' !
#($ ) & * ! µ&
+,$

Slide courtesy of Bill Welsh Slide courtesy of Bill Welsh

Outline De Novo Drug Design

Virtual drug screening General Strategy:
• Ligand-based Given a protein structure
§ 2D structure matching • Build model of binding site
§ 3D structure matching
• Construct molecule that fits model
§ QSAR

De novo drug design

• Models
§ Simulation
§ Knowledge-based
• Construction algorithms
§ Incremental construction
§ Fragment-based
§ Stochastic optimization

5
De Novo Drug Design
General Strategy:
• Given a protein structure
Build model of binding site
• Construct molecule that fits model
De Novo Drug Design
General Strategy:
• Given a protein structure
• Build model of binding site
Construct molecule that fits model

Outline Binding Site Models

Virtual drug screening
• Ligand-based
§ 2D structure matching
§ 3D structure matching
§ QSAR

De novo drug design

Models
§ Simulation
§ Knowledge-based
• Construction algorithms
§ Incremental construction
§ Fragment-based
§ Stochastic optimization
Example courtesy of Joe Corkery

Binding Site Models Outline

Simulation Virtual drug screening
• e.g., GRID • Ligand-based
Dry C3 § 2D structure matching
Knowledge-based § 3D structure matching
• e.g., X-SITE § QSAR

De novo drug design

• Models
§ Simulation
N1 O § Knowledge-based
Predicted Construction algorithms
Binding Site § Incremental construction
Model for
1kp8-1-H-ATP-1-_ § Fragment-based
using GRID § Stochastic optimization
[Goodford85]

6
Incremental Construction Fragment-Based Methods
NH2

O
Protein
N
NH 2 O
O N

O
NH2
Building

1. Docking scaffolds
2. Add reactants (substituents)

Linking Fragments
O
N O
NH 2
N
O NH 2
O

3. Find conformation
4. Minimize & Score
Slide courtesy of Bill Welsh Slide courtesy of Bill Welsh

Stochastic Optimization Summary

Monte Carlo search of chemical space: Virtual drug screening
• Start from initial drug • Ligand-based
• Make random state changes § 2D structure matching
(add/delete/move chemical group) § 3D structure matching
§ QSAR
• Accept up-hill moves with probability
dictated by “temperature” De novo drug design
• Reduce temperature after each move • Models
• Stop after temperature gets very small § Simulation
§ Knowledge-based
• Construction algorithms
§ Incremental construction
§ Fragment-based
§ Stochastic optimization

Discussion