Embed Size (px)
Citation preview
Chimica FarmaceuticaIntroduction to Ligand-Based Drug Design 2
Chimica FarmaceuticaIntroduction to Ligand-Based Drug Design 3
Rino Ragno: Computational Medicinal Chemistry Applications to Epigenetic Targets Inhibitors 409-09-2015
Conformational Search Molecular Docking Simulate Annealing
Molecular Dynamics Graphical Visualization 3-D QSAR Pharmacophore
Ab Initio QM
QSAR, COMBINE, Scoring Functions, Homology Modeling,…..
Rino Ragno: Computational Medicinal Chemistry Applications to Epigenetic Targets Inhibitors 509-09-2015
Basic Research
Target Ident
Target Valid
Hit Ident
Hit to Lead
Lead Opt Preclin Clinical
Trial Prod Diagnos
Drug Design = Computational & Synthesis Tandem
Rino Ragno: Computational Medicinal Chemistry Applications to Epigenetic Targets Inhibitors 609-09-2015
Protein Structure
Unknown Known
Liga
nd S
truc
ture U
nkno
wn
Library Screening De Novo
Know
n
Ligand Based Structure-Based
Ligand-Based
QSAR Phamacophore 3-D QSAR
Structure-Based
ScoringFunctionDocking
COMBINE
Chimica FarmaceuticaIntroduction to Ligand-Based Drug Design 7
Chimica FarmaceuticaIntroduction to Ligand-Based Drug Design 8
Chimica FarmaceuticaIntroduction to Ligand-Based Drug Design 9
Chimica FarmaceuticaIntroduction to Ligand-Based Drug Design 10
Chimica FarmaceuticaIntroduction to Ligand-Based Drug Design 11
Chimica FarmaceuticaIntroduction to Ligand-Based Drug Design 12
Chimica FarmaceuticaIntroduction to Ligand-Based Drug Design 13
Chimica FarmaceuticaIntroduction to Ligand-Based Drug Design 14
Chimica FarmaceuticaIntroduction to Ligand-Based Drug Design 15
Chimica FarmaceuticaIntroduction to Ligand-Based Drug Design 16
The Hansch Equation
Chimica FarmaceuticaIntroduction to Ligand-Based Drug Design 17
The first QSAR equations were based on the observation that partition coefficients, as expressed by log P values, are to some
extent, correlated to certain biological endpoints.
log (1/C) = k1 log P + k2σ + k3
Conc. of compound requiredto produce a standard response
in a given t
Logarithm of the molecule’s partition coefficient
(1-octanol/water)
Hammet Parameter(molecule’s electronic
characteristics)
Chimica FarmaceuticaIntroduction to Ligand-Based Drug Design 18
Chimica FarmaceuticaIntroduction to Ligand-Based Drug Design 19
Chimica FarmaceuticaIntroduction to Ligand-Based Drug Design 20
Identification of Active Ligands
Identification of Suitable Descriptors (molecular fingerprint)
Establish Mathematical Expression Relating Descriptors to Activity
Construction and Validation of the QSAR model
Chimica FarmaceuticaIntroduction to Ligand-Based Drug Design 21
Chimica FarmaceuticaIntroduction to Ligand-Based Drug Design 22
Chimica FarmaceuticaIntroduction to Ligand-Based Drug Design 23
Chimica FarmaceuticaIntroduction to Ligand-Based Drug Design 24
Squared CorrelationCoefficient R2 or r2
Cross-Validated R2 Q2 or q2
∑
∑
=
=
−
−−= N
ii
N
iicalci
YY
YYr
1
2exp,
1
2,exp,
2
)(
)(1
∑
∑
=
=
−
−−= N
ii
N
iipredi
YY
YYq
1
2exp,
1
2,exp,
2
)(
)(1
Chimica FarmaceuticaIntroduction to Ligand-Based Drug Design 25
Squared CorrelationCoefficient R2 or r2
∑
∑
=
=
−
−−= N
ii
N
iicalci
YY
YYr
1
2exp,
1
2,exp,
2
)(
)(1
10 2 ≤≤ r
∑
∑
=
=
−
−−≡−≡= N
ii
N
iicalci
YY
YY
TSSRSS
TSSESSr
1
2exp,
1
2,exp,
2
)(
)(11
Fitting
Chimica FarmaceuticaIntroduction to Ligand-Based Drug Design 26
Cross-validated R2 Q2 or q2
12 ≤≤∞− q∑
∑
=
=
−
−−= N
ii
N
iipredi
YY
YYq
1
2exp,
1
2,exp,
2
)(
)(1
The predictive ability of a model is estimatedusing a reduced set of structural data
CV (Cross-Validation)
N
YYSDEP
N
iipredi∑
=
−= 1
2,exp, )(
Chimica FarmaceuticaIntroduction to Ligand-Based Drug Design 27
Chimica FarmaceuticaIntroduction to Ligand-Based Drug Design 28
Y-Scrambling
Scrambling
Scrambled Y vector
A new model is obtained for such permuted data, R2 and Q2 are then recalculated.This step is repeated for a sufficient number of times (iterations):
a good number being 50 to 100.
Values obtained in the above fashion are compared with the true values obtained for the modelthat was fitted on the real data.
Original X block
A statistical test of prediction tools, in which models are fitted for randomly reorderedproperty/activity values and compared with the model obtained for the actualproperty/activity values.
A new model is obtained for suchpermuted data, R2 and Q2 are thenrecalculated.This step is repeated for a sufficientnumber of times (iterations):
a good number being 50 to 100.
Values obtained in the above fashion arecompared with the true values obtained forthe model that was fitted on the real data.
Original Y vector
Chimica FarmaceuticaIntroduction to Ligand-Based Drug Design 29
N
YYSDEP
N
iipredi∑
=
−= 1
2,exp, )(
N° of predicted compounds
External Test-SetSDEP (Standard Deviation Error of Prediction)
