marcus gastreich a. recent docking score advancements b

Marcus Gastreich

A. Recent Docking Score AdvancementsB. Replacements of Unwanted Molecular Parts

“Anybody up for a push in scoring must at least invest 5 years.“Matthias Rarey, FlexX Original Author

2

A

Recent Advances in Scoring, Hyde.

13-May-11©2011 BioSolve IT 3

Everything is Water. Thales

© Flickr


Basic Concept: Dehydration + H-Bonds

desolvatibindingatom

interacton on

iΔG ΔG ΔGi i

i

* Reulecke et al., ChemMedChem, 2008

∆Gdesolvation

∆Ginteraction


Atom-Based Desolvation

acciconformation‐dependent SAS of atom i

plogPiatom‐based logP contribution

bounddesolvation freeΔG 2.3RT acc acc plogPi i ii

i

i


Atom-Based Interaction

sati number of interactions(degree of saturation)

fsat reference state(saturation of H-bonds in solvent water)

plogPi atom-based logP contribution

interactio bonsat

und free2.3RTΔG sat sat plogP

fi i i

i

i

i273Kfsat = 1

310Kfsat = 0.85


plogP(sp2 oxygen) = ‐1.42

∆G desolvation = 8.09 kJ/mol

plogP(aromatic carbon) = 0.45

∆G desolvation = ‐1.37 kJ/mol

Calibration dataset: 458 small, simple molecules

taken from the Starlist*

21 plogP descriptors used

* Hansch C. et al., American Chemical Society, 1995

Atom-Based logP Contribution


Input/Output - Reality Check

Correct prediction of:

hydrophobic effect (110 J/Å2)

H‐bonds in vacuum (≈ 16 kJ/mol)

H‐bonds in water (‐2 to ‐6 kJ/mol)

affinity loss of unsatisfied H‐bond (6 kJ/mol)

Experimental logPs of small molecules

plogPincrements

fsat

Score reflects affinity of inhibitor to protein

ΔG = ‐RT ln(K)


ΔG = ‐RT ln(K)


ΔG = ‐RT ln(K)

HYDE


Very important:

G = -RT lnK

=> Energy doubled means squared affinity!

Take Home Message

13-May-11©2011 BioSolve IT

10

O8

C6N7

Binding Mode Analysis with HYDE

+∆G contribution

‐∆G contribution

no ∆G contribution

HYDE color code:

receptor aromatic carbons -5.2 kJ/mol

ligand aromatic carbon -2.0 kJ/mol

total desolvation gain -7.2 kJ/mol

NH

CO

Here we have a bond!!

receptor amide N desolv 6.3 kJ/mol

interaction energy -7.4 kJ/mol

ligand aromatic N desolv 6.4 kJ/mol

interaction energy -7.5 kJ/mol

Total energy -2.2 kJ/mol

receptor carbonyl oxygen 8.2 kJ/mol

ligand aromatic oxygen 2.4 kJ/mol

total desolvation cost 10.6 kJ/mol


HYDE: Guiding OptimizationStructure-based design of novel Checkpoint kinase 1 inhibitors:Insights into hydrogen bonding and protein-ligand affinity *

NH

NN

NH

OH

2BRM

O

NN

NH

OH

2BRB

* [Foloppe N. et al., JMedChem 2005]

∆Gexp = -34 kJ/mol∆GHyde = -33 kJ/mol

∆Gexp = -28 kJ/mol∆GHyde = -22 kJ/mol


ΔGHYDE -7 kJ/mol

HYDE Requires Perfection

Crystal structure

∆Gopt = LJpotinter

+ LJpotintraLigand

+ TorsionLigand

+ 2∆GHYDEinteract

+ ∆GHYDEdesolv

Stage 1:

hydrogen bondnetwork

optimization*

ΔGHYDE -29 kJ/mol

* Lippert et al., J.Cheminformatics, 2009

Stage 2:

avoid clashes, optimize

hydrogen bondgeometries

ΔGHYDE -34 kJ/mol

1X8X


Hyde Results – Improved Docking

”Astex Diverse” Testset as defined for the ACS Docking & Scoring competition 2011.

RMS | ≤0.5Å ≤1.0Å ≤1.5Å ≤2.0Å ≤2.5Å-------------------------------------------------rank 1 | 14.0% 46.7% 62.7% 74.7% 80.7% rank ≤ 32 | 43.3% 80.7% 88.0% 90.7% 96.0%rank any | 57.3% 84.7% 92.0% 95.3% 97.3%


Hyde Results – Improved Screening

15.8 13.9FlexX/HYDE

DUD

* Cross et al., JCIM, 2009, 49, 1455ff

*


HYDE – Affinity Assessment

NN

NN

O

Cl

Pyrazole Urea‐Based Inhibitors of p38 MAP Kinase: From Lead Compound to Clinical Candidate *

* Regan et al., JMedChem, 2002



NN

NN

O



NN

N

O

Cl


HYDE – Affinity AssessmentN

N

NN

O


HYDE – Affinity AssessmentN

N

NN

O

O

N

O



NN

NN

O

Cl

NN

NN

O

O

N

O


Detailed Affinity AssessmentMore than a Simple Lipophilic Contact: A Detailed Thermodynamic Analysis of Nonbasic Residues in the S1 Pocket of Thrombin *

NH2 O

N

O

NH

R1

D-Phe ProS1

occupant

2ZDV

2ZC9

2ZDA

2ZFF

2ZF0

* Baum et al., JMB, 2009


2ZFF

2ZF0

2ZC9

HYDE Analysis of Thrombin Inhibitors

∆Gexp ∆GHYDE

-32 -30 -25 -5

-31 -34 -25 -9

-35 -30 -17 -13

-35 -36 -22 -14

-46 -40 -20 -20

D-Phe + Pro S1 occupant

NH2 O

N

O

NH

R1

2ZDA

2ZDV

Correlation: 0.77 (rPearson=0.54)


Conclusions

HYDE ...

is an intuitive and untrained scoring function

scores single atom exchanges correctly

provides a guideline for lead optimization

∆GHYDE -20

∆GHYDE -13


HYDE in LeadIT v2.0

Tutorial: Score Biotin in Streptavidin (1stp)


Tutorial Steps: HYDE

1. Fire up LeadIT

2. Prepare your protein carefully1. Receptor -> Load or Prepare2. (Down)load PDB file 1stp.pdb3. Walk through wizard

3. Hyde! The co-crystallised ligand (the “reference ligand“)1. Go to Scoring -> Assess with Hyde2. Get acquainted with the interface, press HYDE!

4. Analyse and export your graphics into PowerPoint/… using a simple drag&drop1. What does the thumb mean?2. Where are favourable contributions in your molecule?3. What would you do to increase ist activity?

33

BPragmatic Replacement of

Unwanted Parts in Molecules

The Tasks:Bioisostere + Fragment Replacement

34

1


A Molecule Has “Issues“… Physicochemically unwanted:

- Toxic groups- Too hydrophilic/-phobic/…, or no wanted chemistry- Selectivity

Synthetically unwanted:- Hard-to-synthesize moieties- Too expensive building blocks

Strategically unwanted:- Patented scaffolds

Or there is simply something missing to make a molecule (= a link).This is for later…

35


Bioisosteres

Bioisosteres are substituents or groups with similar physical or chemical properties which produce broadly similar biological properties to a chemical compound.

36

metabolically unstable

© wikipedia

© wikipedia


The Famous Viagra Patent Bust

37

The Software Landscape –A surely non‐comprehensive overview.

38 Marcus Gastreich

2

3D Approaches

39 Marcus Gastreich

Feb 2, 2011©2011 BioSolve IT GmbH

BROOD

40Marcus Gastreich

Procedure: 1. Compute shape/electrostatics for a given query; optionally with protein2. Search a library for suitable replacements

Published originally by:• OpenEye (www.eyesopen.com/brood)

Available from: OpenEye Scientific Software

© OpenEye Website


FieldStere

41Marcus Gastreich

Procedure: 1. Compute molecular fields (similar to pharmacophores)2. Search a library for suitable replacements to fit into the fields

(field comparison); optionally with protein information3. Score using molecular fields and/or shape

Published originally by: Cresset Group

Available from: Cresset (www.cresset.com)


SHOP

42Marcus Gastreich

Procedure: 1. Compute interaction energy forcefield (GRID) using probes

(protein inclusion indirectly possible)2. Search a library for suitable replacements

Published originally by:• Fontaine et al., ChemMedChem 2009 4 427

Available from: Molecular Discovery


CAVEAT: A Historical FBLD Helper

43Marcus Gastreich

Procedure: 1. Exit vector definition on query molecule 2. Ranked fragment insertion procedure

(cp. below)

Published originally by:Lauri & Bartlett., J. Comp. Aided Mol. Design 1994, 8, 51-66

Available from: U Berkeleyhttp://www.cchem.berkeley.edu/pabgrp/Data/caveat.html

Some Problems:• Missing simple user interface• Threshold dependence of results• Only fragments are found, but not valid combinations thereof• Synthetic accessibility not considered• No protein inclusion possible• Not all common operating systems supported

O

O

Feb 2, 2011©2011 BioSolve IT GmbH 44Marcus Gastreich

“Scoring“ in CAVEAT

purely geometric ranking

according to Euclidean distance in 4D: distance d torsion 2x angles a1 and a2 formed with connecting path

d

a1

a2Query Vector

Result Vector


MOE Linker Pharmacophores

45Marcus Gastreich

Procedure: 1. Define ‘generic pharmacophores‘ on ligand 2. Search a library of fragments with boundary condition of matching the

generic pharmacophore (similar to CAVEAT vectors)3. If sufficiently good superposition can be obtained, form new bond

Published originally on:• Chemical Computing Group website

Available: In MOE, Chemical Computing Group (www.chemcomp.com)


Roche 2007: ReCore

define “Exit Vectors” search 3D fragment library

Software development by Maass, Rarey (ZBH, Hamburg) and Roche Available from: BioSolveIT (www.biosolveit.de/ReCore)

Maaß, Schulz-Gasch, Stahl Rarey, J. Chem. Inf. Mod., 47, 390-399 (2007)


ReCore: Indexed Searching

ReCore has an “index” to search extremely fast: instead of looking at every single possibility, you jump to the answer in seconds:

The Index needs a pre-generated “descriptor”:

telephone book:the alphabet

an index

ReCore: Vector relationships


ReCore: Results Delivery

By design the best fitting replacements are found first The ranking is according to deviation from perfect vector

overlap

Rank 1

Rank 2

Rank 3

Rank 4ReCore Query


A ReCore Specialty: “3D Shredding”

Advantages:• more possibilities• bridging larger gaps• still respect conformation/torsions seen in root structure

(ideally: experiment/bioactive)

“partial enumeration“

trivial fragmentation1

1

2

2


ReCore Needs at least 2 Vectors

Vectors = EVs and Pharm Features (non-spherical pharmacophores):

Sensible Query Scenarios:2 EV + opt’l: any Pharm Features or EVs1 EV + 1 directed Pharm Feature + opt’l: any Pharm Features

Pharm Features

Exit Vectors


Optional Volumes & Shape

Mimick pocket (pocket spheres, = “allowed area“)

Forbid areas (exclusion spheres, opt‘l: SMARTS)

Refine definitions (inclusion w/ optional SMARTS)

[*]


Three Levels of Synthetic Acc‘y Checks

Retrosynthetic shredding rules = more sensible fragments!

Interactivity: Cut where chemist can synthesize!

Forbid unwanted bond formation!

3. Linking

2. Query definition

1. Fragment creation


Synthetic Access Checks & User Chemistry

Input mol2

O

NH

O

NH

S

NH2

ReasonableFragments

RECAP-ishrules

Filters, eg, corporate

O

NH N

H

O

NBr

OH

F

Index

Interactive Queryingwith chemists being d’accord!

ReCore Search Engine

optional:- shape- binding site- in/exclusions etc.


N

O OA

'C

B

Avoiding Bogus Conformations

1. Torsions within the fragment (typically from expt., or from Corina or the like)

2. Torsions upon forming new single bonds: Link Constraints

N

O SA

'C

B

torsions under control by user, connections can also be forbidden

‘observed’ prior to indexing


Chemical Correctness vs. Creativity

What would you like to see? ( are new torsions)

55Marcus Gastreich

is correct

is wrong

Would you discard solution B if an interesting residue followed?

A

B

NO


Be Sure About What You Really Want!

Chemical Correctness Creativity


ReCore: Proof of Concept (Maass et al.)

HIV Reverse Transscriptase: Question: Results Known Inhibitors?

Recore solutions in yellow + MSC194:MSC194

Recore composite solutions in yellow + known inhibitors:

MOZJUGrank 4

CABVADrank 1 XUCPUG

rank 125

1s9e,Janssen1eet 1ddt


Non-3D Approaches

59Marcus Gastreich


CATS and DOGS

60Marcus Gastreich

Procedure: (DOGS=Des. of Genuine Structures) 1. Use reaction-originating synthons 2. assemble to druglike compounds using mutation and recombination

mimicking reactions virtually

Molecule is represented by shape and pharmacophore features with the CATS (chemically advanced template search) descriptor (= 2D ph4)

Published originally by: Schneider (JCIM 2007 47 2280 & Meth.Mol.Biol. 2011 672 299.)

Developed at: ETH Zurich / Frankfurt University


Topomers

61Marcus Gastreich

Procedure:1. Use reference ligand to create fragments along synthetic route2. Compute steric field properties 3. Compare properties to Topomer library members

(the latter comply with synthetis route boundary conditions)4. Choose the Topomers with highest similarity and assemble molecules

combinatorially.

Published by: Cramer et al. J.Med.Chem. 1999 42 3919

Available from: Tripos (www.tripos.com)


Feature Trees (FTrees) Fragment Spaces

62Marcus Gastreich

Procedure:1. Encode chemistries as Fragment Space

using fuzzy similarities (FTrees)2. Search most similar combination

of chemistry fragments using the Feature Tree descriptor (fragment-by-fragment / buildup)

3. Try to build only those compounds which have a desired minimum similarity with respect to the reference compound

• This way, 1013 virtual compounds can be searched. • The results have a very high likelihood of being accessible.

Selected Success Stories:• Boehringer-Ingelheim (prospective): JCIM 2009 49 270• Pfizer: J.Med.Chem. 2008 51 2468• Arqule (Rojnuckarin at ACS 2009, Salt Lake City)

Published originally by: Rarey and Stahl, JCAMD 2001 15 497.

Available from: BioSolveIT (www.biosolveit.de/FTrees)

Please note: Free KNIME and PipelinePilot modules are available.

N

NH+

NN

NH2NH2

N

O

NH

O

O

OO

Vol.: 44.27Ring cl: 1Profile:

Aro

Don

Acc

Amd

Hyd

Feature Tree

Don

Acc

Aro

Amd

Hyd

Vol.: 33.9Ring cl: 0Profile:

The Concept of FBLD

63 Marcus Gastreich

3

64 Marcus Gastreich

A Striking Analogy to FBLD

Core Replacement

Merging & Linking Growing

Essentially, this requires dealing with multiple input molecules.


Work One or More Fragments Into One Compound

G = -RT lnK

1. Find fragment hit1. Find fragment hit

2. Fill further cavity/-ies;keep LE high

2. Fill further cavity/-ies;keep LE high

3. Arrive at one compound 3. Arrive at one compound

Doubled energy = squared affinity


Grow (or: Evolve)

Step by step elaboration into further subpocketStep by step elaboration into further subpocket


Merge and Link

Connecting multiple fragments (“Linking“)If overlapping: “Merging“Connecting multiple fragments (“Linking“)If overlapping: “Merging“


Why Fragment-Based Design Is Popular

• Fragments explore larger chemical space • combinatorial!• often easier H2L / LO

• Fragments more likely to bind (though more weakly)

• Synthesis driven by target only

• Two weak binders combined can drastically improve affinity:

G = -RT lnK

=> Energy doubled means squared affinity!

ReCore FBLD Applications

71

4

Some Happy Users:Sanofi-Aventis, Merck, Roche, Novartis, Bayer, Arqule, Lilly, AstraZeneca, Genentech, …

72Marcus Gastreich

Linking / Merging: Examples


Fragment Linking – A First, Simple Example

How close do we get to Howard et al.‘s JMC success with Thrombin?

Howard et al., JMC 49 1346 (2006)

IC50: 330M

IC50: 12M

IC50: 1.4nMNNN

N

Cl

SO

O

NH

NH2

OH

MeO

SO

O

NH

OH

MeO

NH

NNN

N

Cl



ReCore Queries with alignment of ligand PDBs 2c93 and 2c90:

This is too easy! Roche do not have simple –CHn– as linkers,so we don’t either…

SO

O

NH

OH

MeO

N

NNN

N

Cl



Can we ‘carve out’ the amine? And still link to the chlorophenyl?

SO

O

NH

OH

MeO

N

NNN

N

Cl



Yes, that’s possible:

IC50: 1.4nM

Found first (= Rank 1):

IC50: ??

published structure:

N

SO

O

NH

OH

MeO

NNN

N

Cl

SO

O

NH

OH

MeO

NH

NNN

N

Cl


... and: Fragment Merging

The idea: Connect multiple known fragment binders where they overlap.


A Tough Merging Example

Thymidylate Synthase Inhibition à la Sunesis (Erlanson et al., PNAS 2000 97 9367)

NS

OO

CO2-

PDB: 1F4EKi: 1.1mM

NH

N NH

N

N

ONH

-O2C

-O2C

O

NH2

mTHFKm: 14mM

“…tosyl group is in roughly the same position and orientation as the benzamide moiety of methylenetetrahydrofolate…”

NS

OO

CO2-

NH

O

COO-

-O2C

24M,further opt’d to 330nM




“…tosyl group is in roughly the same position and orientation as the benzamide moiety of methylenetetrahydrofolate…”

mTHF

Tosylproline




NS

OO

CO2-

NH

N NH

N

N

ONH

-O2C

-O2C

O

NH2N

SO

O

CO2-

NH

O

COO-

-O2C

Published Result

NS

OO

CO2-

NH

COO-

-O2C

ReCore Rank 8after 1sec.

To ensure the result accommodates in the pocket the input ligand envelope shapes were used.


A Growing Example


Growing into a Interaction Pattern

Bcl-2 protein example from Abbott (Oltersdorf et al, Nature 435 (2005) 677).

Let us try to get the -stacking right as in the original finding in ABT-737

2 vectors are needed for the Query:a) Exit Vector b) -interaction ‘Pharm. Feature’

Site 1 Site 2

N N

Cl

O

NH

SO2

NO2

NH N

S

N N

Cl

O

NH

SO2

NO2



Here it is in 3D:



The Result in 2D:

ABT-737

ReCore Solution No.11:

N N

Cl

O

NH

SO2

NO2

NH N

S

N N

Cl

O

NH

SO2

NO2

NH N

SO

O



Solution No. 11 in comparison to ABT-737 (green): Almost identical positions obtained!


Further Resources

88

5


Some Recommended Further ReadingReviews and other publications: Wermuth, C. G. Molecular variations based on isosteric replacements. In The Practice of Medicinal

Chemistry; Wermuth, C. G., Eds.; 1996 202-237. From the medchem bible… Meanwell, N.A., Synopsis of Some Recent Tactical Application of Bioisosteres in Drug Design J. Med.

Chem., 2011, 54 (8), pp 2529–2591 Brilliantly written state-of-the-art overview Sheridan, R. P. The most common chemical replacements in druglike compounds. J. Chem. Inf.

Comput. Sci. 2002, 42, 103–108. Retrospective computer analysis Patani, G. A.; LaVoie, E. J. Bioisosterism: A rational approach in drug design. Chem. Rev. 1996, 96

(8), 3147–3176. Big traditional replacements lists, history, classical vs. non-classical bioisosters Hartenfeller & Schneider Meth. Mol. Biol. 2011 672 299

FBLD computer methods, focus on 2D FBLD methods BioSolveIT Whitepaper on FBLD with FTrees-FS:

http://www.biosolveit.de/whitepapers/archive/WhitepaperMedChem_in_VS.pdf

FBLD Blogs http://practicalfragments.blogspot.com (MedChem focus) http://fbdd-lit.blogspot.com (Literature focus)

FBLD Books (focus on experimental methods) Zartler & Shapiro (Eds.) Fragment-Based Drug Discovery: A Practical Approach (Wiley) Jahnke et al. (Eds.) Fragment-based Approaches in Drug Discovery (Methods and Principle in Medicinal

Chemistry), Wiley-VCH

89Marcus Gastreich

Feb 2, 2011©2011 BioSolve IT GmbH 91

Tutorial Steps: ReCore

Tutorial: Replace -CN(C)Ph- in 4dfr_min

Advanced questions:- Do you find the original fragment?- What happens upon inclusion of

- ligand shape- protein shape

- Import the graphical solution into PPT/Word…

1. Fire up LeadIT2. Go to menu ReCore -> Def. ReCore Query3. “Add“ 4dfr_min 4. Define two exit vectors around the

-C-N(-C)-Ph- group5. Click ReCore6. Browse Results

Thank you so much.

Marcus [email protected]

marcus gastreich a. recent docking score advancements b

Documents