Chemical Data andComputer-Aided Drug Discovery

Mike GilsonSchool of Pharmacymgilson@ucsd.edu

2-0622

Outline

Overview of drug discovery

Structure-based computational methodsWhen we know the structure of the targeted protein

Ligand-based computational methodsWhen we don’t know the protein’s structure

What is a drug?

Small Molecule Drugs

Aspirin

Sildenafil (Viagra)

Glipizide (Glucotrol)

Taxol

Digoxin

Darunavir

Nanoparticles(e.g., packaged small-molecule drugs)

Doxil(liposome package,

extended circulation time,milder toxicity)

Abraxane(albumin-packaged taxol)

http://www.doxil.com/about_doxil.html http://www.abraxane.com/professional/nab-technology.aspx

Biopharmaceuticals

Erythropoietin (EPO)Stabilized variant of a natural protein hormone

Etanercept (Enbrel)Protein with TNF receptor + Ab Fc domainScavenges TNF, diminishes inflammation

http://www.ganfyd.org/index.php?title=Erythropoietin_beta http://en.wikipedia.org/wiki/File:Enbrel.jpg

How are drugs discovered?

Digoxin

Foxglove

Aspirin Taxol

Willow

Pacific Yew

Natural Products

How Aspirin Works

inflammation

platelet activation

Aspirin

platelet inactivation

Biomolecular Pathways and Target SelectionE.g. signaling pathways

http://www.isys.uni-stuttgart.de/forschung/sysbio/insulin/index.html

Target protein

Empirical Path to Ligand DiscoveryCompound library(commercial, in-house,

synthetic, natural)

High throughput screening(HTS)

Hit confirmation

Lead compounds(e.g., µM Kd)

Lead optimization(Medicinal chemistry)

Potent drug candidates(nM Kd)

Animal and clinical evaluation

Compound Libraries

Commercial (also in-house pharma) Government (NIH)

Academia

Computer-Aided Ligand Design

Aims to reduce number of compounds synthesized and assayed

Lower costs

Less chemical waste

Faster progress

HIV Protease/KNI-272 complex

Scenario 1Structure of Targeted Protein Known: Structure-Based Drug Discovery

Protein-Ligand Docking Structure-Based Ligand Design

VDW

Dihedral

Screened Coulombic

+ -

Potential functionEnergy as function of structure

Docking softwareSearch for structure of lowest energy

Energy Determines Probability (Stability)Boltzmann distribution

Ene

rgy

Pro

babi

lity

( )/( ) E x RTp x e

x

Structure-Based Virtual Screening

Compound database 3D structure of target(crystallography, NMR, modeling)

Virtual screening(e.g., computational docking)

Candidate ligands

Experimental assay

Ligands

Ligand optimizationMed chem, crystallography, modeling

Drug candidates

Fragmental Structure-Based Screening

“Fragment” library 3D structure of target(crystallography, NMR, modeling)

Fragment docking

Compound design

http://www.beilstein-institut.de/bozen2002/proceedings/Jhoti/jhoti.html

Experimental assay and ligand optimizationMed chem, crystallography, modeling Drug candidates

Physics-Based

Knowledge-Based

Potential Functions for Structure-Based DesignEnergy as a function of structure

Physics-Based PotentialsEnergy terms from physical theory

Van der Waals interactions (shape fitting)Bonded interactions (shape and flexibility)Coulombic interactions (charge-charge complementarity)Hydrogen-bonding

Common Simplifications Used in Physics-Based Docking

Quantum effects approximated classically

Protein often held rigid

Configurational entropy neglected

Influence of water treated crudely

Proteins and Ligand are Flexible

+

Ligand

Protein

Complex

D Go

Binding Energy and Entropy

Unbound states

Bound states

l 3n lnbound FreeG RT E EK RTD

EFree

EBound

Energy part Entropy part

/

/

2

6

Bound

Free

RE

RTE

TeK

e

Structure-Based DiscoveryPhysics-oriented approaches

WeaknessesFully physical detail becomes computationally intractableApproximations are unavoidableParameterization still required

StrengthsInterpretable, provides guides to designBroadly applicable, in principle at leastClear pathways to improving accuracy

StatusUseful, far from perfectMultiple groups working on fewer, better approxs

Force fields, quantumFlexibility, entropyWater effects

Moore’s law: hardware improving

Knowledge-Based Docking Potentials

Histidine

Ligandcarboxylate

Aromaticstacking

Probability Energy

( )/( ) E r RTp r e

( ) ln ( )E r RT p r

Boltzmann:

Inverse Boltzmann:

Example: ligand carboxylate O to protein histidine N

1. Find all protein-ligand structures in the PDB with a ligand carboxylate O2. For each structure, histogram the distances from O to every histidine N3. Sum the histograms over all structures to obtain p(rO-N)4. Compute E(rO-N) from p(rO-N)

“PMF”, Muegge & Martin, J. Med. Chem. 42:791, 1999Knowledge-Based Docking Potentials

A few types of atom pairs, out of several hundred total

Atom-atom distance (Angstroms)

( )( )

( )prot lig vdw type ij ijpairs ij

E E E r

Nitrogen+/Oxygen- Aromatic carbons Aliphatic carbons

Structure-Based DiscoveryKnowledge-based potentials

WeaknessesAccuracy limited by availability of dataAccuracy may also be limited by overall approach

StrengthsRelatively easy to implementComputationally fast

StatusUseful, far from perfectMay be at point of diminishing returns

Limitations of Knowledge-Based Potentials

1. Statistical limitations (e.g., to pairwise potentials)

2. Even if we had infinite statistics, would the results be accurate? (Is inverse Boltzmann quite right? Where is entropy?)

r1 r2 r10…

10 bins for a histogram of O-N distances

rO-N

rO-C

100 bins for a histogram of O-N & O-C distances

rO-N

e.g. MAP Kinase Inhibitors

Using knowledge of existing inhibitors to discover more

Scenario 2Structure of Targeted Protein Unknown: Ligand-Based Drug Discovery

Why Look for Another Ligand if You Already Have Some?

Experimental screening generated some ligands, but they don’t bind tightly

A company wants to work around another company’s chemical patents

An high-affinyt ligand is toxic, is not well-absorbed, etc.

Ligand-Based Virtual Screening

Compound Library Known Ligands

Molecular similarityMachine-learning

Etc.

Candidate ligands

Assay

Actives

OptimizationMed chem, crystallography, modeling

Potent drug candidates

Sources of Data on Known LigandJournals, e.g., J. Med. Chem.

Some Binding and Chemical Activity Databases

PubChem (NIH) pubchem.ncbi.nlm.nih.govChEMBL (EMBL) www.ebi.ac.uk/chemblBindingDB (UCSD) www.bindingdb.org

BindingDBwww.bindingdb.org

Finding Protein-Ligand Data in BindingDB

e.g., by Name of Protein “Target”

e.g., by Ligand Draw Search

Sample Query ResultsBindingDB to PDB

PDB to BindingDB

Download data inmachine-readableformat

Sample Query Results

Machine-Readable Chemical FormatStructure-Data File (SDF)

PDB Format Lacks Chemical BondingSDF Format Defines Chemical Bonds

There are Many Other Chemical File FormatsInterconvert with Babel

Chemical SimilarityLigand-Based Drug-Discovery

Compounds(available/synthesizable)

Compare with known ligands

SimilarTest experimentally

Different

Don’t bother

Chemical FingerprintsBinary Structure Keys

Molecule 1

Molecule 2

phenyl

methyl

ketone

carboxy

late

amidealdehyd

e

chlorin

e

fluorine

ethylnaphthyl

S-S bond

alcohol

…

Chemical Similarity from FingerprintsTanimoto Similarity or Jaccard Index, T

0.25U

ITN

N

NI=2Intersection

NU=8Union

Molecule 1

Molecule 2

Hashed Chemical FingerprintsBased upon paths in the chemical graph

1-atom paths: C F N H S O2-atom paths: F-C C-C C-N C-S S-O C-H3-atom paths: F-C-C C-C-N C-N-H C-S-O

C S-O etc.

Each path sets a pseudo-random bit-pattern in a very long molecular fingerprint

Maximum Common Substructure

Ncommon=34

Potential Drawbacks of Plain Chemical Similarity

May miss good ligands by being overly conservative

Too much weight on irrelevant details

Scaffold Hopping

Zhao, Drug Discovery Today 12:149, 2007

Identification of synthetic statins by scaffold hopping

Abstraction and Identification of Relevant Compound Features

Ligand shape

Pharmacophore models

Chemical descriptors

Statistics and machine learning

+ 1

Bulky hydrophobe

Aromatic

5.0 ±0.3 Å3.2 ±0.4 Å

2.8 ±0.3 Å

Pharmacophore ModelsΦάρμακο (drug) + Φορά (carry)

A 3-point pharmacophore

Molecular DescriptorsMore abstract than chemical fingerprints

Physical descriptorsmolecular weightchargedipole momentnumber of H-bond donors/acceptorsnumber of rotatable bondshydrophobicity (log P and clogP)

Topologicalbranching indexmeasures of linearity vs interconnectedness

Etc. etc.

Rotatable bonds

A High-Dimensional “Chemical Space”Each compound is at a point in an n-dimensional space

Compounds with similar properties are near each other

Descriptor 1

Descriptor 2

Des

crip

tor 3

Point representing a compound in descriptor space

Statistics and Machine LearningSome examples

Partial least squares

Support vector machines

Genetic algorithms for descriptor-selection

Summary

Overview of drug discovery

Computer-aided methodsStructure-basedLigand-based

Interaction potentialsPhysics-basedKnowledge-based (data driven)

Ligand-protein databases, machine-readable chemical formats

Ligand similarity and beyond

Mike Gilson, School of Pharmacy, mgilson@ucsd.edu, 2-0622

Activities and Discussion Topics

BindingDB: Advil Machine-readable format, Binding activities

PDB/BindingDB2ONY at PDB BindingDB Substructure search Related data

Similarity search

Combined computational approaches(physics + knowledge)-based docking potentials(ligand + structure)-based computational discovery

Other data-driven methods where it may be hard to get enough statistics

Validation of computational methods

Protein-ligand databases: getting data and assessing data quality

Drug Discovery Pipeline(One Model)

Target identification

Target validation

Assay development

Animal Pharmacokinetics,

Toxicity

Phase I Clinical(safety, metab, PK)

Phase II Clinical(efficacy)

Phase III Clinical(comparison with existing therapy)

Lead optimization

Lead compound

(ligand) discovery

Muegge J. Med. Chem. 49: 5895, 2006

Updated Knowledge-Based PMF Potential