P5

1
Antonio Llinàs Martí
ADMET Prediction Fiction or Reality?
Antonio Llinàs Martí The Pfizer Institute for
Pharmaceutical Materials Science University of
Cambridge
2
Antonio Llinàs Martí
Is ADMET important?
3
Antonio Llinàs Martí
Why to predict Physico-Chemical properties?
4
Antonio Llinàs Martí
ADMET predictive models
?Linear models MLR (Multiple Linear
Regression) PLS (Partial Least Squares) PCR
(Principal Components Regression) ?Non Linear
models ANN (Artificial Neural Networks) RF
(Random Forest) SVM (Support Vector Machines)
5
ADMET predictive models
Antonio Llinàs Martí
Building a Model
Good Model R2 1 RMSE 0 BIAS 0
6
Antonio Llinàs Martí
ADMET predictive models
Building Good Data

• David Palmer, John Mitchell
• Unilever Centre For Molecular Informatics,
  University of Cambridge

7
Antonio Llinàs Martí
ADMET predictive models
Artificial Neural Networks
Artificial Neural Networks (ANNs) have been used
to distinguish drug-like and non-drug-like
molecules using a substructural analysis Jain
1998. So and Karplus So 1997 used
electrostatic and steric properties at grid
points for feeding a genetic artificial neural
network in order to develop a QSAR model.
Running the network consists of Forward pass
the outputs are calculated and the error at the
output units calculated. Backward pass The
output unit error is used to alter weights on the
output units. Then the error at the hidden nodes
is calculated (by back-propagating the error at
the output units through the weights), and the
weights on the hidden nodes altered using these
values. For each data pair to be learned a
forward pass and backwards pass is performed.
This is repeated over and over again until the
error is at a low enough level (or we give up).
8
Antonio Llinàs Martí
ADMET predictive models
Random Forest
A Decision Tree Forest is an ensemble
(collection) of decision trees whose predictions
are combined to make the overall prediction for
the forest. A decision tree forest is similar to
a TreeBoost model in the sense that a large
number of trees are grown. However, TreeBoost
generates a series of trees with the output of
one tree going into the next tree in the series.
In contrast, a decision tree forest grows a
number of independent trees in parallel, and they
do not interact until after all of them have been
built.
The sinking of the Titanic
Source "Report on the Loss of the 'Titanic'
(S.S.)" (1990), British Board of Trade Inquiry
Report (reprint), Gloucester, UK Allan Sutton
Publishing
9
Antonio Llinàs Martí
ADMET predictive models
Support Vector Machines
The SVM analysis attempts to find a 1-dimensional
hyperplane (i.e. a line) that separates the cases
based on their target categories. There are an
infinite number of possible lines two candidate
lines are shown above. The question is which line
is better, and how do we define the optimal line.
Rather than fitting nonlinear curves to the data,
SVM handles this by using a kernel function to
map the data into a different space where a
hyperplane can be used to do the separation.
The concept of a kernel mapping function is very
powerful. It allows SVM models to perform
separations even with very complex boundaries
such as this one.
10
Multiple Linear Regression
Antonio Llinàs Martí
ADMET predictive models

• Log.S 0.07nHDon (/-0.018) - 0.21TPSA
  (/-0.033) 0.11MAXDP (/-0.022) - 0.22n.Ct
  (/-0.019) - 0.29KierFlex (/-0.032) - 0.59SLOGP
  (/0.036) - 0.26ATS2m (/-0.026) 0.25RBN
  (/-0.033)

David Palmer, John Mitchell. Unilever Centre
For Molecular Informatics, University of Cambridge
11
Random Forest
Antonio Llinàs Martí
ADMET predictive models
RMSE(te)0.69 R2(te)0.89 Bias(te)-0.04
RMSE(tr)0.27 R2(tr)0.98 Bias(tr)0.005
RMSE(oob)0.68 R2(oob)0.90 Bias(oob)0.01
David Palmer, John Mitchell. Unilever Centre
For Molecular Informatics, University of Cambridge
12
Antonio Llinàs Martí
Problems with the actual literature data bases
i. Egregious errors in reporting data and
references
ii. Poor data quality and/or inadequate
documentation
procedures
Pontolillo, J. and Eganhouse, P., U.S. Department
of Interior. U. S. Geological Survey.
Water-Resources Investigations Report 01-4201.
Reston. Virginia. 2001
13
Antonio Llinàs Martí
Problems with the actual literature data bases
i. Egregious errors in reporting data and
references

• S. E. Adams, J. M. Goodman, R. J. Kidd, A. D.
  McNaught, P. Murray-Rust, F. R. Norton, J. A.
  Townsend and C. A. Waudby Org. Biomol. Chem.
  2004, 2, 3067-3070.

14
Antonio Llinàs Martí
Problems with the actual literature data bases
i. Egregious errors in reporting data and
references
Citation Analysis
1259. C. Lee, W. Yang, R. G. Parr, Phys. Rev. B,
1988, 37, 785.9 DUPLICATES FOUNDNEAR
MATCHES1382. C. Lee, W. Yang, R. G. Parr, Phys.
Rev. B, 1988, 37, 785-789.3008. C. Lee, W. Yang,
R. G. Parr, Phys. Rev., 1988, 785-788 .4199. C.
Lee, W. Yang, R. Parr, Phys. Rev. B, 1988, 37,
785.6006. C. Lee, W. Yang, R. G. Parr, Phys.
Rev. B, 1998, 37, 785. 9038. C. T. Lee, W. T.
Yang, R. G. Parr, Phys. Rev. B, 1988, 37,
785.9125. C. Lee, W. Yang, R. G. Parr, Phys.
Rev. B, 1993, 37, 785.11481. C. Lee, W. Yang, R.
G. Parr, Phys. Rev. B, 1988, 37, 785-789.10742.
C. T. Lee, W. T. Yang, R. G. Parr, Phys. Rev. B,
1988, 37, 785-789.
Bruce Russell, Jonathan Goodman (Unilever
Centre For Molecular Informatics, University of
Cambridge)
15
Antonio Llinàs Martí
Problems with the actual literature data bases
i. Egregious errors in reporting data and
references
Citation Analysis
1259. C. Lee, W. Yang, R. G. Parr, Phys. Rev. B,
1988, 37, 785.9 DUPLICATES FOUNDNEAR
MATCHES1382. C. Lee, W. Yang, R. G. Parr, Phys.
Rev. B, 1988, 37, 785-789.3008. C. Lee, W. Yang,
R. G. Parr, Phys. Rev., 1988, 785-788 .4199. C.
Lee, W. Yang, R. Parr, Phys. Rev. B, 1988, 37,
785.6006. C. Lee, W. Yang, R. G. Parr, Phys.
Rev. B, 1998, 37, 785. 9038. C. T. Lee, W. T.
Yang, R. G. Parr, Phys. Rev. B, 1988, 37,
785.9125. C. Lee, W. Yang, R. G. Parr, Phys.
Rev. B, 1993, 37, 785.11481. C. Lee, W. Yang, R.
G. Parr, Phys. Rev. B, 1988, 37, 785-789.10742.
C. T. Lee, W. T. Yang, R. G. Parr, Phys. Rev. B,
1988, 37, 785-789.
Bruce Russell, Jonathan Goodman (Unilever
Centre For Molecular Informatics, University of
Cambridge)
16
Antonio Llinàs Martí
Problems with the actual literature data bases
i. Egregious errors in reporting data and
references
Pontolillo, J. and Eganhouse, P., U.S. Department
of Interior. U. S. Geological Survey.
Water-Resources Investigations Report 01-4201.
Reston. Virginia. 2001
17
Antonio Llinàs Martí
Problems with the actual literature data bases
i. Egregious errors in reporting data and
references
ib. Data errors
Pontolillo, J. and Eganhouse, P., U.S. Department
of Interior. U. S. Geological Survey.
Water-Resources Investigations Report 01-4201.
Reston. Virginia. 2001
18
Antonio Llinàs Martí
ii. Poor data quality and/or inadequate
documentation
procedures
Temperature Solubility g/l Reference
25 25 2.132 896.2 1 2
1 Oliveri-Mandala, E. (1926), Gazzetta Chimica
Italiana 56, 896-901 2 Ochsner, A. B., Belloto,
R. J., and Sokoloski, T. D. (1985), Journal of
Pharmaceutical Sciences 74, 132-135
19
Antonio Llinàs Martí
Solubility definition
Huge range of definitions
Saq
S0
Ksp
Kinetic Solubility Thermodynamic
Solubility Equilibrium Solubility Apparent
Solubility Ionic Solubility Solubility
product Intrinsic Solubility Aqueous
Solubility Standard Solubility ...
S0
Ko
Sw
ST
K0
20
Antonio Llinàs Martí
Solubility definition
Solute
21
Antonio Llinàs Martí
Solubility definition
Solubility- Concentration of a compound in a
saturated solution when excess solid is
present Aqueous Solubility- Concentration of a
compound in a saturated solution of pure water
when excess solid is present. Thermodynamic
Solubility- Solubility when the compound in
solution is at equilibrium with the solid
form. Kinetic Solubility Solubility at the
time when an induced precipitate first appears in
a solution Intrinsic solubility- Of an
ionisable compound is the thermodynamic
solubility of the free acid or base form (Horter,
D, Dressman, J. B., Adv. Drug Deliv. Rev., 1997,
25, 3-14)
22
Antonio Llinàs Martí
Process of dissolution
23
Factors Influencing Solubility
Antonio Llinàs Martí
Temperature Salinity pH Dissolved organic
matter (DOM) Co-solvents Crystallinity Polymorphis
m
- If the solute is subject to acid/base reactions
then pH is vital in determining water
solubility.
24
Factors Influencing Solubility
Antonio Llinàs Martí
Temperature
Generally as T increases, solubility increases
for solids. The influence of temperature on water
solubility can be quantitatively described by the
van't Hoff equation as  ln Csat -?H/(RT)
Const.
25
Factors Influencing Solubility
Antonio Llinàs Martí
Salinity
As salinity increases, the solubility of neutral
organic compounds decreases (activity coefficient
increases)   Ks Setschenow salt constant
(depends on the compound and the salt) The
addition of salt makes it more difficult for the
organic compound to find a cavity to fit into,
because water molecules are busy solvating the
ions.
26
Factors Influencing Solubility
Antonio Llinàs Martí
pH
pH effect depends on the structure of the solute.
If the solute is subject to acid/base
reactions then pH is vital in determining water
solubility. The ionized form has much higher
solubility than the neutral form. The apparent
solubility is higher because it comprises both
the ionized and neutral forms. The intrinsic
solubility of the neutral form is not affected.
27
Factors Influencing Solubility
Antonio Llinàs Martí
Dissolved organic matter (DOM)
DOM increases the apparent water solubility
Solubility in water in the presence of DOM is
given by the relation Csat,DOM Csat (1
DOMKDOM)  DOM concentration of DOM in
water, kg/L KDOM DOM/water partition
coefficient Again, the intrinsic solubility of
the compound is not affected.
28
Factors Influencing Solubility
Antonio Llinàs Martí
Co-Solvents
Co-solvents increase the solubility of
hydrophobic organic chemicals. Co-solvents can
completely change the solvation properties of
water
Solubility increases exponentially as cosolvent
fraction increases.
fv is the volume fraction of co-solvent sic is
the slope term, which depends on the both the
cosolvent and solute
29
Factors Influencing Solubility
Antonio Llinàs Martí
Crystallinity
Crystallinity decreases the apparent solubility
Crystallinity () Apparent Solubility M 35 C
88.6 36.7 20.8 3.50 x 10-3 4.39 x 10-3 5.27 x 10-3
30
Factors Influencing Solubility
Antonio Llinàs Martí
Polymorphism
Crystallising into different crystal forms will
result in different melting points and
solubilities
Crystalline Form MP Apparent Solubility M 25 C
I II III 68 58 42 5.70 x 10-3 6.30 x 10-3 7.40 x 10-3
31
Solubility measurements
Antonio Llinàs Martí
Classical MethodShake Flask Method
? Many published variations of this method
32
Antonio Llinàs Martí
Diclofenac An Example
CheqSol is a new method developed by
33
Antonio Llinàs Martí
Diclofenac
Precipitation ? As soon as pptate is detected
titrant addition stops ? pH keeps going up
because AH is removed from solution and A- reacts
with H to replace the AH lost ? The solution,
at this point, is SUPERSATURATED NOT IN
EQUILIBRIUM
34
Antonio Llinàs Martí
Diclofenac
Dissolution ? After pptation is confirmed an
aliquot of base is added ? pH goes down because
AH (solid) is brought back in solution, AH
(ston), generating A- and H ? The solution, at
this point, is SUBSATURATED NOT IN EQUILIBRIUM
35
Antonio Llinàs Martí
Diclofenac
? We continue Chasing equilibrium until a
specified number of crossing points have been
reached ? A crossing point represents the moment
when the solution switches from a saturated
solution to a subsaturated solution no change in
pH, gradient zero, no re-dissolving nor
precipitating. SOLUTION IS IN EQUILIBRIUM
Si 1.53 0.15 mg/ml
36
Diclofenac Characterisation
Antonio Llinàs Martí
NO MATCH !!!!
37
Diclofenac Characterisation
Antonio Llinàs Martí
MATCH !!!!
38
Diclofenac Characterisation
Antonio Llinàs Martí
Diclofenac Acid C2/c polymorph Polyhedron
(1993), 12, 1361
39
Diclofenac Characterisation
Antonio Llinàs Martí
?
Thanks to John Davies for solving the X-ray
structure of this crystal
40
Diclofenac Characterisation
Antonio Llinàs Martí
DSC- MP 263.4 C TGA- Pentahydrate
DSC- MP 267.4 C TGA- Anhydrous
DSC- MP 180.5 C TGA- Anhydrous
41
Diclofenac Solubility (25C, I 0.15 M)
Antonio Llinàs Martí
Si 1.49 0.09 mg/ml
Si 1.53 0.15 mg/ml
Si 1.47 0.12 mg/ml
42
Diclofenac Complete
Antonio Llinàs Martí
Powder XRD- ? Single Crystal XRD- EA- BAD MP
(DSC)- 267.4 C TGA- Sodium Salt
Anhydrous Solubility 1.528 0.15 mg/ml
Powder XRD- NEW Single Crystal XRD- SOLVED
Sodium salt PentahydrateP2(1) EA- OK MP (DSC)-
263.4 C TGA- Sodium Salt Pentahydrate Solubility
1.472 0.09 mg/ml
Powder XRD- SIKLIH01 Single Crystal XRD- NO EA-
OK MP (DSC)- 180.5 C TGA- Diclofenac Acid
Anhydrous Solubility 1.488 0.12 mg/ml
43
Antonio Llinàs Martí
Conclusions

• Predictive ADMET is in its infancy
• Models are not improving
• Actual databases are no good bad quality data,
  no
• diverse enough
• Need of high quality data to build reliable
  databases
• Need of standardization. Same conditions, same
  definition, characterisation, and statistical
  treatment
• Solubility Intrinsic, 25 C, I 0.15 M (KCl),
  purity of starting material gt99.5 , Solid
  characterisation.

44
Antonio Llinàs Martí
Acknowledgments
- Sirius Analytical Instruments Ltd.
Karl Box
- Unilever Solubility Team
Prof. Robert Glen (Director)Dr. Jonathan Goodman
(Group Leader)Dr. John Mitchell (Group
Leader)Dr. Antonio LlinàsDavid Palmer
- University of Cambridge
To ALL of YOU