Focused Virtual Screening a presentation by Dr David Lloyd Trinity College Dublin Daylight EuroMUG 2004 Lead discovery in the Human Estrogen Receptor
Focused Virtual Screening
a presentation by Dr David Lloyd
Trinity College DublinDaylight EuroMUG 2004
Lead discovery in the Human Estrogen Receptor
1592
750 AD
Biochemistry in TCD – largest Department in Country
Significant research output
Centre for Structural Biology and Molecular Design
building on PRTLI and SFI investment in structural biologybuilding on PRTLI and SFI investment in structural biology
Molecular Design Group
Established 2004
Ireland’s first protein X-ray facility
Integrated Drug Discovery
Chemistry Biology
Computation
Structure Based Design – looking in the ER
Structure Based Design – looking in the ER
Curr Med Chem 2003, Frontiers Med Chem 2005 (in press)
Significance of ER
• Estrogens regulate cell growth, differentiation & development of reproductive tissues in men and women.
• Maintain bone density preventing osteoporosis.
• Exerts anti-atherosclerotic effects which lowers Cholesterol levels.
• Involved in many CNS effects (Parkinson's) and implicated in Alzheimer's.
ER as a target
• 60% of primary breast cancers contain ER- alpha
• Estrogens are mitogenic for ER-positive breast cancer cells.
Target: Target: EstrogenEstrogen Receptor Receptor ��
Structure Based Design – Docking in Nuclear Receptors
Docking AlgorithmsDocking Algorithms Scoring FunctionsScoring Functions
Structure Based Design – Docking in Nuclear Receptors
FREDFRED
FlexXFlexX
Discover3Discover3
eHitseHits
In houseIn house
PLPPLP
ChemScoreChemScore
CScoreCScore
PLCPLC
Structure Based Design – looking in the ER
Building on knowns : using receptor structural knowledge– semi-rational design
Traditional scaffold hopping –human de novo rational design
J Med Chem2001
Anti Cancer Drug Design
2001
Traditional scaffold hopping –human de novo rational designComputer-enhanced!!
Benzoxepin antiestrogens
N
aminoalkylation
aminoalkylation
BBr3
PhZnCl
Pd(PPh3)4
boronic acids
Pyr.HCl
PyHBr3H+n-BuLi
OMe
Br
1110
987
para CN (19)para Me (15)
meta Me (16)para Cl (17)meta NO2 (18)
para OMe (12)ortho OMe (13)meta OMe (14)R2 =
21-2820
R1=
NN
N
O
N
6 7-115
Suzuki Route
Heck Route
4
3
R2
O
O
N
O
R1O
O
HO
Br
O
MeO
O
HO
R2
2
O
HO
O
MeO
Br
O
MeO
O
HO
OMe
O
O
Computer-enhanced human de novo rational design
Ortho-ring substitutionis tolerated - meta is not -
elcectic binding mode
Computer-enhanced human de novo rational design
J Med Chem2004
Haystack built from 880 ‘drug-like’ compounds from WDI
�40 Cox-2 inhibitors�40 Estrogen Receptor Modulators �40 Histamine ‘modulators’
Active ‘needles’ introduced from a separate validated ligand set
Let the computer decide : Virtual Screening
Virtual Screening
vHTS – Performance Measures – Validation
Enrichment = hit rate observed in subsethit rate in database (random)
Enrichment Subset Size (%)
1 5 10 15 20
Ligands
10
50
100
150
200
Max Actives
10
40
40
40
40
Best Possible
Value
25
20
10
6.7
5.0
e.g. 1% sampled = 10 compounds.Subset - 10 actives = hit rate of 10/10 = 1.0, Hitrate in database is 40/1000= 0.04 : enrichment = 1 / 0.04 = 25
Target Database
Docking Protocol
Rescoring
Generation of Hits
Active set of compounds for development
-Remove waters & Calculate centre of bound ligand.-Use multiple structures
PreProcessing
• Samples search space and generates a set of binding poses for each ligand conformer.
• Docked positions have their respective hydrogen bond lengths optimized to allow for refinement of the final structure.
• CF (Complementarity Function) evaluates fit
• Ranks these modes/ligand positions
• Provides a numerical score allowing for ‘hit’identification
In-house docking protocol
Actual Hits Retrieved
0
5
10
15
20
25
30
35
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
% Sample Database
% H
its R
etri
eved
F_Score
G_Score
PMF_Score
D_Score
ChemScore
Xscore
PLP_Score
Fresno
Screenscore
Hammerhead
Theoretical Maximum HitsRetreived
Chemscoreperforms best of scoring functions.
Accounts for:Hydrogen bond contacts, Lipophiliccontacts, entropic penalty.
G-score focuses on hydrogen bonding interactions only for example.
Getting it right – Scoring Functions
-38.208COX20080-94.16(Drug_401)
-39.083ESTR0067-94.24(Drug_472)
-39.112Drug_61-94.43(Drug_751)
-39.176Drug_259-94.5(Drug_476)
-39.624Drug_421-94.53(Drug_466)
-39.874Drug_265-95.72(ESTR0085)
-40.123ESTR0068-96.07(ESTR0024)
-40.154Drug_315-96.18(ESTR0045)
-40.389Drug_257-96.33(Drug_161)
-40.687Drug_217-96.64(ESTR0043)
-40.991Drug_249-97.1(Drug_163)
-41.031Drug_416-100.14(ESTR0034)
-41.635Drug_823-101.03(ESTR0046)
-41.647Drug_219-102.62(Drug_474)
-42.169Drug_440-103.16(Drug_160)
-42.209Drug_353-103.35(Drug_633)
-42.485Drug_344-103.42(Drug_159)
-44.427ESTR0072-104.85(Drug_158)
-46.568ESTR0079-107.73(ESTR0025)
FREDCHEMSCOREName_IDSybyl6.91CHEMSCOREName_ID
Getting it right – Scoring Functions
Hit Retrieval
0
5
10
15
20
25
30
35
40
1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91 97
% Database Screened
% H
it R
etri
eved
In House Protocol
FlexX
FRED
Best Value
Getting it right – early method validation
Getting it right – Ligand Pre Processing
Virtual High-Throughput Screening
<1 sec per compound – rigid/rigid system
CorinaCorina OmegaOmega StergenStergenRubiconRubicon QuacPacQuacPac
0.4740_RUBICON_LEVEL1
0.5740_CATALYST_LEVEL1
0.4840_OMEGA_LEVEL1
0.5440_CORINA_LEVEL1
X
0.6440_RUBICON_LEVEL2
0.6240_CATALYST_LEVEL2
0.6640_OMEGA_LEVEL2
0.6440_CORINA_LEVEL2
Quac-X
0.6940_RUBICON_LEVEL3
0.6940_CATALYST_LEVEL3
0.7140_OMEGA_LEVEL3
0.6940_CORINA_LEVEL3
Quac-X-10 Confs
0.7440_RUBICON_LEVEL4
0.6340_CATALYST_LEVEL4
0.7040_OMEGA_LEVEL4
0.6440_CORINA_LEVEL4
Quac-Ster-X
0.7440_RUBICON_LEVEL5
0.7340_CATALYST_LEVEL5
0.6940_OMEGA_LEVEL5
0.6940_CORINA_LEVEL5
Quac-Ster-X-10 Confs
Getting it right – Ligand Pre Processing
Random screening – 40 actives in 1000 –each active returns a score – the bigger the difference between the active and inactive scores, the better the method
Preprocessing can increase the cutoff value for ligand consideration – reducing the subset we must consider in order to find our active ligands.
Getting it right – Ligand Pre Processing
6.58.7512.516.2520LEVEL1_RUBICON
67.59.1712.517.5LEVEL1_CATALYST
78.751013.7522.5LEVEL1_OMEGA
1011.881516.2522.5LEVEL1_CORINA
Avgenrichment4321Subset %
11.513.1314.213.7510LEVEL2_RUBICON
55.6256.668.7515LEVEL2_CATALYST
9.511.2514.218.7522.5LEVEL2_OMEGA
7.511.2518.32025LEVEL2_CORINA
910.62514.1618.7522.5LEVEL3_RUBICON
910.62513.332025LEVEL3_CATALYST
9.511.251521.2525LEVEL3_OMEGA
9.511.8751521.2525LEVEL3_CORINA
1113.7514.1718.7522.5LEVEL4_RUBICON
6.56.8758.331017.5LEVEL4_CATALYST
9.510.62513.3316.2520LEVEL4_OMEGA
9.511.251521.2525LEVEL4_CORINA
1213.7516.6622.525LEVEL5_RUBICON
911.2513.3318.7525LEVEL5_OMEGA
89.37514.172025LEVEL5_OMEGA
9.511.251521.2525LEVEL5_CORINA
18 purchased and assayed
Does it really work ?– Validate, Validate, Validate
Compound Number IC50 in MCF-7 MTTMDG-ER-001 8.23E-07MDG-ER-002 8.00E-06MDG-ER-003 2.02E-05MDG-ER-004 5.59E-04MDG-ER-005 6.06E-04TAMOXIFEN 5.51E-06
Screen ligands, prepare ranked hitlist
cluster hits – 20 clusters
5 Hits 5 Hits µµµµµµµµm rangem range4 Chemical Classes4 Chemical Classes3 novel Chemotypes3 novel Chemotypes
MW 450MW 450--550550LOGP 4.8LOGP 4.8--6.56.5
What else do we need?
Familial scoring functions
Validation Validation Validation
Flexible systems – dynamics in dockingChemical intelligence in fragment assembly
Tiered Discovery – integration of technologies
System Simulation
Acknowledgements
The ER collaboratorsDr Mary Meegan – School of Pharmacy TCDDr Vladimir Sobolev – Weismann Institute, IsraelProf James Sexton – Trinity Centre for High Performance ComputingProf Clive Williams – Biochemistry TCDDr Daniela Zisterer – Biochemistry TCDDr Amir Khan – Biochemistry TCD
The workers The facilitatorsAndy Knox Dermot FrostYidong YangGiorgio CartaValeria OnnisGeorgia Golfis