parafrag a new tool for bioisoster searching - cepos insilico · harald mauser: parafrag cepos...
TRANSCRIPT
ParaFragA New Tool for Bioisoster Searching
Harald Mauser
2Harald Mauser: ParaFrag Cepos Symposium 2010
Local PropertiesMolecular properties are not isotropic
Max
Min
Classical MEP Quantum mechanical MEP
ClSCH3
O
O
Implemented in most pharmacophore searchand docking algorithms
see work on σ-hole bonding by Clark and Politzer (eg J Mol Model 2007)
MEP
3Harald Mauser: ParaFrag Cepos Symposium 2010
The Four Local Properties Derived from quantum-mechanical calculations*
Local Ionization Energy (IEL)Lewis Donor Properties → electrophilic processes:
Local Electron Affinity (EAL)Lewis Acceptor Properties → nucleophilic processes
Local Polarizability (αL)Dispersion Interactions → London forces
Molecular Electrostatic Potential (MEP): Electrostatic Interactions
(*) ParaSurf, Cepos InSilico Ltd, Ryde, UK (http://www.ceposinsilico.com)
4Harald Mauser: ParaFrag Cepos Symposium 2010
The ParaFrag Pharmacophore Local properties & surfaces
1. 3D structure
2. IEL surface
3. IEL extremes
Factor-Xa inhibitor taken from X-Ray structure 2vwo (K. Zbinden et al. Eur J Med Chem. 2009, 44, 2787)
5Harald Mauser: ParaFrag Cepos Symposium 2010
-80
-60
-40
-20
0
20
40
0 500 1000 1500 2000 2500 3000 3500 4000 4500
no. surface pointsM
EP v
alue
Calculation of the ParaFrag PharmacophoreFast statistical approach
1. Property distribution (Example MEP)
6Harald Mauser: ParaFrag Cepos Symposium 2010
Calculation of the ParaFrag PharmacophoreFast statistical approach
2. Consider extremes above (robust Median + 2*MAD) or below (robust Median –2*MAD)
3. Identify representative centroids
keep only
• one representative per cluster• top 6 extremes per property
-80
-60
-40
-20
0
20
40
0 500 1000 1500 2000 2500 3000 3500 4000 4500
no. surface points
MEP
val
ue
7Harald Mauser: ParaFrag Cepos Symposium 2010
A First Application: Scaffold HoppingRetrospective Analysis of FactorXa Inhibitors*
NH
N NH
N
N
O
O
NH
O NH
N
N
O
OMEP
surface
IEL
surface
IEL
critical points
QUERY TARGET
Pharmacophore match = 0.59(0.47 automated exit vector matching on conformer library)
(*) Jakobi, A.-J.; Mauser, H.; Clark T. J. Mol. Model. 2008, 14, 547-558.
8Harald Mauser: ParaFrag Cepos Symposium 2010
• In house data with activity and structural information
• Activity sensitive to electronic properties
• Dataset biased to active compounds
• Challenging as in-actives show high similarity to actives
• Training for:
- Separation of actives and inactives (full molecule similarity comparison)
- Bioisosteres
- Comparison of molecular fragments across different series
ParaFrag: Training on In-house DatasetOutline
9Harald Mauser: ParaFrag Cepos Symposium 2010
The ParaFrag Pharmacophore Local properties & surfaces
1. 3D structure
2. IEL surface
3. IEL extremes
Factor-Xa inhibitor taken from X-Ray structure 2vwo (K. Zbinden et al. Eur J Med Chem. 2009, 44, 2787)
10Harald Mauser: ParaFrag Cepos Symposium 2010
The ParaFrag Pharmacophore in MOEExtension of the MOE Pharmacophore Schemes
11Harald Mauser: ParaFrag Cepos Symposium 2010
Key interactions coded by local surface properties Extremes & Interactions
1bfgy
Co-crystallized factor-Xa inhibitor (PDB-ID: 2vwo; K. Zbinden et al. Eur J Med Chem. 2009, 44, 2787)
…. H-bond
…. vdW
…. non-classical
MEP
IEL
EAL
POL
max min
min
max
12Harald Mauser: ParaFrag Cepos Symposium 2010
Key interactions coded by local surface propertiesExample: π-stacking
1bfgy
IEL extremes for π interactions
max
min
Co-crystallized factor-Xa inhibitor (PDB-ID: 2vwo; K. Zbinden et al. Eur J Med Chem. 2009, 44, 2787)
13Harald Mauser: ParaFrag Cepos Symposium 2010
Key interactions coded by local surface properties Example: non-classical interactions
1bfgy
MEP
IEL
EAL
POL
max min
min
max
…. vdW
…. non-classical
14Harald Mauser: ParaFrag Cepos Symposium 2010
ParaFrag: Training on In-house Dataset Learnings
• Definition of defaults for ParaFrag pharmacophores
– Feature radius: 1.0 Å (all properties)
– Partial matches min. six features
– Shape filter (optional): 3.0 Å exterior volume
– Exit vector alignment (bioisosteres): 0.2 Å tolerance (EV1 & EV2)
• Revision of statistical identification of extremes
– correct representation of local symmetries
– correlation with observed interaction patterns (visual control)
15Harald Mauser: ParaFrag Cepos Symposium 2010
ParaFrag – Final Validation with Full moleculesIn-house validation sets for virtual screening
Kinase
9600 decoys, 96 actives
after stereoisomer enumeration 18062 : 114
Enzyme A
10000 decoys, 100 actives
after stereoisomer enumeration 22026 : 140
Serin Protease
6500 decoys, 66 actives
after stereoisomer enumeration 12400 : 94
16Harald Mauser: ParaFrag Cepos Symposium 2010
ParaFrag – Final Validation with Full moleculesWorkflow
Pipeline Pilot Corina Omega Rocs ParaFrag
EON
(Re-)Scoring by ET_Combo Pharmacophore matching
& (Re-)Scoring by
ParaFrag-Score*
Selecting single conformers by shape,
Aligning (needed by ParaFrag)
& Scoring by ShapeTanimoto
(additional: ComboScore)
top5000
5000
Enumeration of stereoisomeres
& Protonation (Moka pH 6-8)
top
(*) ParaFrag-Score = No. Matched Features/ No. Features Query Molecule
17Harald Mauser: ParaFrag Cepos Symposium 2010
ParaFrag – Final Validation with Full molecules Enrichment – Enzyme A
Query structure 1: Query structure 2:
18Harald Mauser: ParaFrag Cepos Symposium 2010
ParaFrag – Final Validation with Full molecules Enrichment – Serine Protease
Query structure 1: Query structure 2:
19Harald Mauser: ParaFrag Cepos Symposium 2010
ParaFrag – Final Validation with Full MoleculesConclusions
• ParaFrag performs comparable to established virtual screening (VS) tools
• Recognition of surface extremes reliable for structurally diverse molecules
• Defaults from in-house training set proved to be robust across different targets
Now we are ready for Bioisosters
20Harald Mauser: ParaFrag Cepos Symposium 2010
ParaFrag – Final ValidationBioisoster Searching
• Scope:– Transfer knowledge from virtual screening – Reproduce results known from literature– Identify novel, putative bioisosters (visual inspection)
• Dataset: – Static Bioisoster database
• Query Fragments:– Carboxylic acid ester
– Aniline– Pyridine
21Harald Mauser: ParaFrag Cepos Symposium 2010
The ParaFrag PharmacophoreBioisoster Search
Molecular fragment with surface extremesStatic Bioisoster database:
- fragments derived from CSD using Recore*
- experimental conformations
- filter for small fragments (MW ≤ 150)
- max. 3 exit vectors (EV)
- approx. 70,000 template fragments in total
- link-atoms saturated with methyl for QM calculation (automated process)
– surface extremes for fragments stored in MOE db high number of extremes to account
for directionality of interactions(*) Maass, P.; Schulz-Gasch, T.; Stahl, M.; Rarey, M. J. Chem. Inf. Model. 2007, 47, 390 -399.
22Harald Mauser: ParaFrag Cepos Symposium 2010
Example Bioisoster SearchingEster
known
OO
Rreference R
NN
O
R
N ON
R
NN N
S
R
ON
newR
NN
R
N
N
N
R
O
RN
partial matches!
23Harald Mauser: ParaFrag Cepos Symposium 2010
Example Bioisoster SearchingSimple Aniline
referenceR
NH2
knownR
NN
H
N
R
NN
H
new
R
NN
N H
R
NN N
S
R
NO
N
partial match(electron-rich π-system)
24Harald Mauser: ParaFrag Cepos Symposium 2010
Example Bioisoster SearchingSimple Pyridine
referenceR
N
R
NN
R'
RN
N NN
known
new
R
CH3
RN
S
R
O
R
O
RN
R
OH
partial match(π-system)
25Harald Mauser: ParaFrag Cepos Symposium 2010
Critical Considerations
• Recognition of the local symmetry of surface properties is challenging
use spherical harmonics for detailed analysis
• Instabilities in accessing the compute-server
employ new MOE WSDL/SOAP technology
• Long response time for MOE pharmacophore searches
code not yet optimized for speed
(CPU-intensive)
26Harald Mauser: ParaFrag Cepos Symposium 2010
Conclusions
• ParaFrag is an innovative approach for bioisoster searching
• Could be valuable for rescoring in specific casescomparable performance to established VS tools over all targets investigated
• Easy to use graphical interface for interaction hot spot analysis
• Robust defaults for different search strategies established
• User has full access to all parameters refinement of pharmacophores
shape filtering
27Harald Mauser: ParaFrag Cepos Symposium 2010
Acknowledgements
StudentsLennart Anger Arjen Jakobi
Chemical Technology and InnovationAlex Alanine
Caterina Bissantz
Wolfgang Guba
Jérôme Hert
Bernd Kuhn
Olivier Roche
Tanja Schulz-Gasch
Daniel Stoffler
Chemical Computing GroupGuido Kirsten (MOE Interface)
Wolfram Altenhofen
Alex Clark (WSDL/SOAP)
Computer-Chemie-Centrum, University Erlangen-NürnbergTim Clark
Cepos InSilico
We Innovate Healthcare