Computer Aided Drug Design

Qin Xu

Contact

• 4-221 Life Science Building • Email: xuqin523@sjtu.edu.cn • Tel: 34204348(O) • Website:

http://cbb.sjtu.edu.cn/~qinxu/teaching.htm

Readings • 分子模拟与计算机辅助药物设计

– 魏冬青等编著，上海交通大学出版社 • 计算机辅助药物分子设计

– 徐筱杰等编著，化学工业出版社 • 计算机辅助药物设计导论

– 叶德泳著,化学工业出版社 • Others

– Textbook of Drug Design and Discovery, Third Edition, by Povl Krogsgaard-Larsen, published by Oxford University Press

– Computational Medicinal Chemistry for Drug Discovery， by Patrick Bultinck, et. al，published by Marcel Dekker， 2003

– Molecular Modeling for Beginners，by Alan Hinchliffe，published by John Wiley and Sons，2002

Outline • Introduction and Case Study • Drug Targets

– Sequence analysis – Protein structure prediction – Molecular simulation

• Drug Design – Combinatorial library – De novo Drug Design – Pharmacophore – QSAR

• Molecular Docking

What is a drug?

• Defined composition with a pharmacological effect

• Regulated by the Food and Drug Administration (FDA)

• Most drugs are small molecules, and the interactions they make with proteins determine their effects, and toxicity, to the human body

Sources of Drugs • Small Molecules

– Natural products • fermentation broths(发酵液） • plant extracts • animal fluids (e.g., snake venoms蛇毒)

– Synthetic Medicinal Chemicals • Project medicinal chemistry derived • Combinatorial chemistry derived

• Biologicals – Natural products (isolation) – Recombinant products – Chimeric or novel recombinant products

Drug Discovery and Drug Development

• Drug Discovery – Concept, mechanism, assay, screening, hit

identification, lead demonstration, lead optimization

– In Vivo proof of concept in animals and concomitant demonstration of a therapeutic index

• Drug Development – Begins when the decision is made to put a

molecule into phase I clinical trials

Issues in Drug Discovery • Hits and Leads 成药性

– Is it a “Druggable” target? • Resistance 抗药性 • Pharmacodynamics 药效学 • Delivery - oral and otherwise 投药方式 • ADMET (absorption, distribution, metabolism,

and excretion - toxicity ) • Patentability

Drug Discovery Processes

Molecular Biological Hypothesis (Genomics)

Chemical Hypothesis

Physiological Hypothesis

Primary Assays Biochemical Cellular Pharmacological Physiological

Sources of Molecules Natural Products Synthetic Chemicals Combichem Biologicals

+ Initial Hit Compounds Screening

Drug Discovery Processes - II

Initial Hit Compounds

Secondary Evaluation - Mechanism Of Action - Dose Response

Initial Synthetic Evaluation - analytics - first analogs

Hit to Lead Chemistry - physical properties -in vitro metabolism

First In Vivo Tests - PK, efficacy, toxicity

Drug Discovery Processes - III

Lead Optimization Potency Selectivity Physical Properties PK Metabolism Oral Bioavailability Synthetic Ease Scalability

Pharmacology

Multiple In Vivo Models

Chronic Dosing

Preliminary Tox

Development Candidate (and Backups)

`> 10,000,000 compounds

1 drug

>1,000 “hits”

6 drug candidates

10-15 years $300 to >$800 million

preclinical clinical Phase I III

12 “leads”

Drug Discovery Pipeline

• The time from conception to approval of a new drug is typically 10-15 years

• The vast majority of molecules fail along the way

• The estimated cost to bring to market a successful drug is now $800 million!! (Dimasi, 2000)

活性化合物

先导化合物

药物候选物 药物

Drug Discovery Disciplines • Medicine • Physiology/pathology • Pharmacology • Molecular/cellular biology • Automation/robotics • Medicinal, analytical,and combinatorial

chemistry • Structural and computational chemistry • Bioinformatics

Why we need computers?

• Clinical trials are most expensive part of the pipeline – if failure can be predicted before this point, it saves time and money

A Little History of Computer Aided Drug Design

• 1960’s - review the target - drug interaction • 1980’s- Automation - high throughput target/drug selection • 1980’s- Databases (information technology) - combinatorial libraries • 1980’s- Fast computers - docking • 1990’s- Fast computers - genome assembly - genomic based target selection • 2000’s- Vast information handling - pharmacogenomics

CADD in various stages of drug discovery

Some introductions in this class

先导化合物设计与优化

药物候选物 随机筛选 1，000-2，000 个化合物

临床前研究

临床研究

市场

理论计算、分子模拟 计算机辅助药物设计

2~3年

2~3年 2~3年

2~3年

3~4年

Modeling Docking

Molecular Simulation

Combinatorial library De novo Drug Design Pharmacophore QSAR

Outline • Introduction and Case Study • Drug Targets

– Sequence analysis – Protein structure prediction – Molecular simulation

• Drug Design – Combinatorial library – De novo Drug Design – Pharmacophore – QSAR

• Molecular Docking

Drug Targets • Receptor, enzyme, ion channel, and

nucleic acid which can act with drug. • Therapeutic Target Database (TTD) http://bidd.nus.edu.sg/group/ttd/ttd.asp A database to provide information about the known and explored

therapeutic protein and nucleic acid targets, the targeted disease, pathway information and the corresponding drugs directed at each of these targets.

Also included in this database are links to relevant databases containing information about target function, sequence, 3D structure, ligand binding properties, enzyme nomenclature and drug structure, therapeutic class, clinical development status.

Therapeutic Target Database (TTD)

Drugs and targets • One drug on one target （FDA standard） • Multi target agents

– One drug on multiple target – Side effects

• Drug combinations – Complexed biological metabolism – Interference between drugs – Cock tail to overcome Drug resistance

• Nature derived drugs, traditional medicine

Types of Drug Combinations • Pharmacodynamically (药代动力学)

– synergistic drug combinations due to anti-counteractive actions

– synergistic drug combinations due to complementary actions

– synergistic drug combinations due to facilitating actions

– additive drug combinations – antagonistic drug combinations – potentiative drug combinations – reductive drug combinations

Successful drug developments

• HIV-1 Protease Inhibitors in the market: – Inverase (Hoffman-LaRoche, 1995) – Norvir (Abbot, 1996) – Crixivan (Merck, 1996) – Viracept (Agouron, 1997)

Drug discovery today 2, 261-272 (1997)

HIV-1 virus

Nat

ure

Med

icin

e 5,

740

- 74

2 (1

999)

Integrase inhibitors

Cocktail therapy of AIDS

Serendipity in Drug Discovery

• Surprised effect of one drug designed for other proposes – Tamoxifen (birth control and breast cancer) – Viagra (hypertension 高血压 and erectile

dysfunction 勃起功能障碍) – Salvarsan (Sleeping sickness and syphilis 胎传性梅毒)

– Interferon-α (hairy cell leukemia 多毛细胞白血病 and Hepatitis C丙肝)

Sequence Alignment • Dynamic programming sequence alignment

– Needleman/Wunsch global alignment – Smith/Waterman local alignment – Linear and affine gap penalties

• Improved algorithms • Heuristic algorithms

– FASTA – BLAST – CLUSTAL

Protein structure prediction • Sequence

– From RNA sequence – Protein sequencing

• Structure

– Secondary – Tertiary

• Function

– Activity, specificity – Binding

Protein structure prediction

• Secondary structure prediction – Distribution of amino acids – Featured sequenceFeatured domain

• Tertiary structure prediction

– The 3D structure of target proteins – Structure of active site, binding site,

possible conformational changes – Molecular docking, molecular dynamics

simulations

Protein 3D structure prediction

• Homology Modeling – Homolog template – Similar sequence similar secondary

structure similar featured domain similar backbone similar 3D structure with optimized side chains

• Threading method – Fold recognition method – Long distance homology protein

• Ab initio prediction – Sequence Structure – Conformer search – Energy minimization

Rosetta

MOE, MODELLER, SWISS-MODEL

% structure overlap the fraction of equivalent residues (Cα atoms within 3.5Å each other).

Sequence identity and Accuracy

Sequence identity and applications

Molecular Simulations

• Molecular Dynamics, Langevin Dynamics – Protein structure prediction – Molecular docking – Predictive ADMET

• Monte-Carlo – Combinatorial library – De novo drug design – Molecular docking

Applications of Molecular Simulations

• Search of conformations – Drug molecules – Target proteins – Binding complex

• Energetic determination – Structural optimization – Binding energy – Free energy simulations

Concepts of Molecular Simulations • Statistical Mechanics

– Phase space – Ensembles: N, V, E and N, P, T

• Scales of simulations – Quantum Mechanical – Molecular Mechanical and force fields – Coarse grained

• Related methods – Boundary condition – Potential truncation – Constraints

Forces calculation and Scales of simulations

http://www.sciencedirect.com/science/article/pii/S0079642509000565

Common workflow in running MD simulations 1. Model Generation. （建模） 2. Energy minimization. （能量最小化） 3. Heating-up the system to the

desired temperature. （加温） 4. Equilibration. （平衡） 5. Production run. （采样计算） 6. Analysis. （分析）

Examples of biomedical researches in computer-aided Drug design

• Much more gigantic system – 104~108 atoms

• Much more complicated – target protein – drug ligand – Solution, membrane environment

• Two examples – nicotinice acetylcholine receptor, nAChR – influenza virus proton channel M2.

nAChR

Ruo-Xu Gu, 2012, Structural Bioinformatics Studies of Two Ion Channel Proteins.

Interactions between JN403 and three different nAChRs

the VdW interactions between JN 403 and the residues at the binding cavity the long rang electrostatic interactions between JN403 and the extracellular domain of the receptors are responsible for the ligand’s selectivity for different nAChR subtypes

Ruo-Xu Gu, 2012, Structural Bioinformatics Studies of Two Ion Channel Proteins.

The correlation of motions of the receptor extracellular domain induced

by agonists, varenicline

Cα of α subunits

Ruo

-Xu

Gu,

201

2, S

truct

ural

Bio

info

rmat

ics

Stud

ies

of T

wo

Ion

Cha

nnel

Pro

tein

s.

The correlation of motions of a~h

Ruo

-Xu

Gu,

201

2, S

truct

ural

Bio

info

rmat

ics

Stud

ies

of T

wo

Ion

Cha

nnel

Pro

tein

s.

Possible binding sites of the allosteric modulators

Ruo-Xu Gu, 2012, Structural Bioinformatics Studies of Two Ion Channel Proteins.

Influenza viral infection to host cell

Behrens G., Stoll M., Kamps B.S., et al., Pathogenesis and immunology. Printed by Druckhaus Sud GmbH & Co. KG, D-50968 Cologne, 2006, 19.

Binding to M2 Channel

• Two sites • Three

paths for ligands

Ruo-Xu Gu, 2012

Structural Bioinformatics Studies of Two Ion Channel Proteins.

Binding at P-site

Rotation of the rimantadine molecule during the MD simulation explained the NMR result.

• amantadine (金刚烷胺 ) bound

• Rimantadine (金刚乙胺 ) bound

Ruo-Xu Gu, 2012

Structural Bioinformatics Studies of Two Ion Channel Proteins.

Block of M2 channel

Free energy of four binding states

rimantadine in the M2-lipid environment

P: thermodynamic preferred (hard to bind, but stably bound)

S: kinetic preferred (easy to come, easy to go)

Ruo-Xu Gu, 2012, Structural Bioinformatics Studies of Two Ion Channel Proteins.

Virtual combinatorial library • Fragment screening （分子片段枚举） • RECAP （分子片段化与重组）

– RECAP analysis – RECAP synthesis

• BREED（配体繁殖） – Crossover by Genetic algorithm – superposition

Terms for combinatorial library

Functional group

R-group

Leaving group

Attachment point

Scaffold

Reagent

Reactive atoms

De novo Drug Design

• Not de novo, but by database searching – Fragment-based – Descriptor-based

• Descriptors and Molecular Docking – Scaffold replacement – Fragment evolution – Link & Merge

Fragment-based Fragments

(databases)

Target active site

(3D structure)

Docking Evaluation Link & Merge

Descriptor-based

Fragments

(databases)

Target active site

(3D structure)

Pharmacophore Query & Docking

Receptor surface calculation

Pharmacophore & Link features

Pharmacophore

Geometric arrangement of functional groups of the ligand that are required for “activity”

A pharmacophore is a spatial arrangement of atoms or functional groups believed to be responsible for biological activity

• Pharmacophore-based alignment

Pharmacophore of KZ7088

7 points pharmacophores

The ligand KZ7088 in the active site of SARS-CoV Mpro

A pharmacophore scheme is a collection of functions that define the meaning, appearance and methods of calculation of ligand annotation points and their attached labels. The scheme defines how each ligand in the searched database is annotated. The default scheme is PCH (Polarity-Charge-Hydrophobicity).

Label Definition Don Hydrogen bond donors, including tautomeric donors. Acc Hydrogen bond acceptors, including tautomeric acceptors. Cat Cations, including resonance cations. Ani Anions, including resonance anions. Hyd Hydrophobic areas. Aro Aromatic centers.

Quantitative Structure-Activity Relationships (QSAR)

• Mathematical Models Found a quantitative relationship

Molecular Parameters

log P σ

MR MLP

…

Biological activity IC50 Ki MIC Permeation …

Statistical tool

The key of QSAR

Descriptor Choice

Statistical Model construction

Fingerprint

{ is-aromatic, has-ring, has-C }

{ has-ring, has-C }

{ has-C, has-N, has-O }

{ is-aromatic, has-ring, has-C, has-N, has-F }

If a universe of features U = { is-aromatic, has-ring, has-C, has-N, has-O, has-S, has-P, has-halogen } there are 28=256 possible fingerprints Tanimoto coefficient between fingerprints X and Y is defined to be: # features in intersect(A,B) / # features in union(A,B)

3D-Molecular Field Calculation

Mol3

Mol1Mol2

PLS Bio = cte + a*S001 + b*S002 + ..... + m*S999 +n*E001 + ..... + z*E999

Bio S001 E001 E999....S002 S999....

Interpretation of CoMFA results

Compounds with low activity

Compounds with high activity

3D-QSAR contour

COMFA Electrostatic

3D-QSAR Parameters

• Traditional CoMFA fields – Steric fields, Lennard-Jones potential

– Electrostatic fields, Coulomb potential

E = r + r

r - 2

r + r rj

k = 1

natoms probe k

ij

12probe k

jkε∑ •

•

6

E = q q

rjprobe k

jkk = 1

natoms •

•∑

ε

More 3D-QSAR Parameters

• Additional fields in CoMFA – Interaction energies with functional groups (GRID software)

– hydrophobic field (HINT software)

– Molecular Lipophilicity Field (CLIP software)

Biological activity parameters

– IC50 (50% inhibiting concentration), – Ki (inhibitory constant), – MIC (minimum inhibitory concentration), – Permeation

Statistical models • Linear (quantitative model)

– LR (linear regression) – MLR (multi-linear regression) – PCR (principle component regression) – PLS (partial least square) – SVR (support vector regression)

• Nonlinear (qualitative model) – SVM (support vector machine) – Bayesian statistics

∑∑

−

−−= 2

2exp

)()(

1 Rmeancalc

calc

yyyy

拟合效果 • 相关系数R，

• 均方根偏差RMSE

• Fisher检验值F

1)(

1 RMSE2

exp

−−

−−= ∑

knyycalc

2

2

)1()1(

RkknRF

−−−

=

calcy 为计算活性； expy 为实验活性； meany

为计算活性平均值； n为样本个数； k 为变量个数

Evaluation of QSAR models

Statistical value for 3D-QSAR •For the step of crossvalidation:

•Cross-validated correlation coefficient, q2. •Optimal number of components, N.

•For the final model: •Squared of correlation coefficient, r2. •Standard error of estimate, s. •Residuals. •F values.

Molecular Docking • Multistep procedure

– Prescreening: ligands or binding site – Pose optimization – Binding energy and evaluations

• Representations of the receptor – Atomic – Surface – Grid

• Scoring schemes – Force field-based: Electrostatic, VDW – Empirical – Knowledge-based scoring functions

Molecular Docking • Systematic docking algorithms

– conformational search methods – fragmentation methods – database methods.

• Random or stochastic algorithms – Monte Carlo methods (MC) – Genetic Algorithm methods (GA) – Tabu Search methods

• Simulation methods – Molecular dynamics – Local minimization

• Flexible protein docking – MD and MC methods – rotamer libraries – protein ensemble grids – soft-receptor modeling

Docking

• Prediction of ligand conformation and orientation (or posing) within a targeted binding site – accurate structural modeling – correct prediction of activity

• Free energy of binding (∆G)

Goals of Docking 1) Characterize binding site - make an image of binding site with

interaction points 2) Orient ligand into binding site - grid search - descriptor mapping - energy search 3) Evaluate strength of the interaction ∆G bind= ∆G complex – (∆Gligand +

∆Gtarget)

Multi-step Process • First, find the poses of small molecules in the

active site • Using simple scoring functions evaluating

compound fits on the basis of calculations of approximate shape and electrostatic complementarities.

• Pre-selected conformers are often further evaluated using more complex scoring schemes – More detailed treatment of electrostatic and van der Waals interactions – inclusion of at least some solvation or entropic effects – And other empirical terms

Molecular Representation • Three basic representations of the

receptor: – atomic, surface and grid

• Atomic representation – only used in conjunction with a potential energy function – often only during final RANKING procedures

• Surface-based docking – Typically used in protein–protein docking – Connolly’s surface

• Potential energy – store information about the receptor’s energetic contributions on

grid points so that it only needs to be read during ligand scoring. – two types of potentials: electrostatic and van der Waals

Standard Potential Energy • Electrostatic potential energy:

• Van der Waals interaction – Lennard-Jones 12-6 function

Docking Models • System: ligand, protein, and solvent

molecules. • Solvent molecules

– they are normally excluded from the problem. – Or implicitly modeled in the scoring functions as a way to address

the solvent effect.

• Rigid-body approximation – very popular in early – treats both the ligand and the receptor as rigid and explores only

the 6 degrees of translational and rotational freedom.

• A more common approach – modeling ligand flexibility while assuming a rigid protein receptor,

therefore considering only the conformational space of the ligand.

Docking algorithms • Systematic docking algorithms • Random or stochastic algorithms • Simulation methods • Flexible protein docking

Systematic Docking • Explore all the degrees of freedom in a

molecule – conformational search methods – fragmentation methods – database methods.

• Conformational search methods – All possible combinations have been generated and evaluated

• Fragmentation methods – “the place-and-join approach” – “the incremental approach”

• Database – use libraries of pre-generated conformations

Stochastic Algorithms • Three basic types of

– Monte Carlo methods (MC) – Genetic Algorithm methods (GA) – Tabu Search methods. (tabu = forbidden)

• Monte Carlo methods – Sampling Boltzmann probability function – Significant advantage over molecular dynamics methods (MD)

• Genetic algorithms – GAs start from an initial population of different conformations of the

ligand with respect to the protein.

Simulation Methods • Two major types exist

– molecular dynamics (MD) – pure energy minimization methods

• Molecular dynamics methods – difficulties in navigating a rugged hypersurface of a biological

system and crossing high-energy barriers – the problem in sampling the conformational space within a feasible

simulation period

• Energy minimization methods – simplex, gradient methods, conjugate-gradient methods, etc. – rarely used as a stand-alone search technique in docking because

only local minima can be reached

Flexible Protein Docking • MD and MC methods, rotamer libraries,

protein ensemble grids, soft-receptor modeling

• MD and MC: – docking simulations with a fully flexible target are currently not feasible.

• Rotamer libraries – Represent the protein conformational space as a set of experimentally

observed and preferred rotameric states for each side chain

• Ensemble of protein conformations – Multiple protein structures from crystal structures, NMR, or calculations

• The Soft-receptor modeling approach – combines the information of several different protein conformations to

generate a single “energy weighted average” grid

Scoring Functions • Lack of a suitable scoring function, both

in terms of speed and accuracy, is the major bottleneck in docking

• Generally able to predict binding free energies within 7–10 kJ/mol

• Three major classes – Force field-based – Empirical – Knowledge-based scoring functions

Force Field-based Scoring • Generally the sum of two energies:

– interaction energy between the receptor and the ligand – the internal energy of the ligand – Lennard-Jones + electrostatic terms

• D-Score, G-Score, GoldScore, and the AutoDock 3.0 scoring function

• Complete molecular mechanical force-fields: – AMBER (Assisted Model Building and Energy Refinement) – CHARMM (Chemistry at HARvard Macromolecular Mechanics) – GROMOS(Groningen Molecular Simulation System) – OPLS (Optimized Potentials for Liquid Simulations)

Empirical Scoring Functions • Designed to reproduce experimental data and are

based on the idea that binding energies can be approximated by a sum of several individual uncorrelated terms.

• Experimentally determined binding energies and sometimes a training set of experimentally resolved receptor–ligand complexes are used to determine the coefficients for the various terms by means of a regression analysis.

• Dependence on the experimental data set used in the parameterization process (non versatile and non transferable).

Knowledge-Based Scoring Functions

• Trying to implicitly capture binding effects that are difficult to model explicitly.

• Typically, very simple atomic interactions pair potentials

• Based on the frequency of occurrence of different atom–atom pair contacts and other typical interactions in large datasets of protein–ligand complexes of known structure.

Capabilities and Limitations • Predict known protein-bound poses with

averaged accuracies of about 1.5–2 A with success rates in the range of 70–80%.

• Significant improvement beyond this range seems for now unachievable, even with the inclusion of receptor flexibility.

• Source of problems – Scoring function is the major limiting factor. – solvent effect and the direct participation of water molecules in protein–ligand

interactions, – Limited resolutions of most crystallographic targets – Protein flexibility

• Recent trends – Inclusion of solvation and rotational entropy contributions – development of search algorithms more able to describe and efficiently sample the

conformational space of the protein–ligand system, within a flexible-target paradigm

Software

Docking – Cancer Drug Design • Ligand (analog-based)

• Target (structure-based)

Protein-Ligand Interactions

Quiz

quantitative? ligand or receptor?

energy or geometry?

local or global?

QSAR

Pharmacophore

Molecular Docking De novo

drug design

In developments

quantitative? ligand or receptor?

energy or geometry?

local or global?

QSAR yes ligand energy or geometry global

Pharmacophore no ligand or

receptor geometry local

Molecular Docking yes ligand &

receptor energy or geometry

global or local

De novo drug

design yes ligand &

receptor energy or geometry local