EPA/600/R-03/033
March 2003
VERIFICATION AND VALIDATION OF THE SPARC MODEL
By
S.H. Hilal and S.W. Karickhoff
Ecosystems Research Division
U.S. Environmental Protection Agency
Athens, Georgia
and
L.A. Carreira
Department of Chemistry
University of Georgia
Athens, Georgia
National Exposure Research Laboratory
Office of Research and Development
U.S. Environmental Protection Agency
Research Triangle Park, NC 27711
-------
DISCLAIMER
The United States Environmental Protection Agency through its Office of Research and
Development partially funded and collaborated in the research described here under assistance
agreement number 822999010 to the University of Georgia. It has been subjected to the Agency
peer and administration review process and approved for publication as an EPA document.
ABSTRACT
SPARC (SPARC Performs Automated Reasoning in Chemistry) chemical reactivity models were
validated on more than 5000 ionization pKas (in the gas phase and in many organic solvents including
water as a function of temperature), 1200 carboxylic acid ester hydrolysis rate constants (as a function
of solvent and temperature), 350 Ei/2 reduction potential (as a function of solvents, pH and temperature),
and 250 gas phase electron affinities. Physical properties have been developed and validated using more
than 8000 physical property data points on many properties such as vapor pressure (as a function of
temperature), boiling point (as a function of pressure), solubility, activity coefficients, Henry's constant
and Kow (as a function of solvent and temperature), etc. However, the true validity of the SPARC property
models is the ability of the SPARC basic models to be extended to calculate numerous chemical/physical
properties (as a function of solvent, temperature, pressure, pH, etc.) without modification or extra
parameterization to any of the SPARC basic models.
-------
FOREWORD
Recent trends in environmental regulatory strategies dictate that EPA will rely heavily on
predictive modeling to carry out the increasingly complex array of exposure and risk assessments
necessary to develop scientifically defensible regulations. In response to this need, researchers at ERD-
Athens have developed a predictive modeling system SPARC (SPARC Performs Automated Reasoning
in Chemistry) that calculates a large number of physical and chemical properties from pollutant
molecular structure across all classes of industrial organic chemicals. SPARC execution involves the
classification of molecular structures and the selection and execution of appropriate "mechanistic"
models, such as induction, resonance, and field effects to quantify reactivity. The basic mechanistic
models in SPARC were designed and parameterized to be portable to any type of chemistry or organic
chemical structure. This expanded prediction capability allows one to choose, for exhaustive
validating, the reaction parameters for which large and reliable data sets do exist to validate against.
The SPARC models have been validated on more than 12,000 data points for many properties. The
verification and validation of the SPARC models will be presented in this report.
Rosemarie C. Russo, Ph.D.
Director
Ecosystems Research Division
Athens, Georgia
in
-------
SUMMARY
The major differences among behavioral profiles of molecules in the environment are
attributable to their physicochemical properties. For most chemicals, only fragmentary knowledge
exists about those properties that determine each compound's environmental fate. A chemical-by-
chemical measurement of the required properties is not practical because of expense and because trained
technicians and adequate facilities are not available for measurement efforts involving thousands of
chemicals. In fact, physical and chemical properties have only been measured for about 1 percent of the
approximately 70,000 industrial chemicals listed by the U.S. Environmental Protection Agency's Office
of Prevention, Pesticides and Toxic Substances [1].
Although a wide variety of approaches are commonly used in regulatory exposure and risk
assessments, knowledge of the relevant chemistry of the compound in question is critical to any
assessment scenario. For volatilization, sorption and other physical processes, considerable success has
been achieved in not only phenomenological process modeling but also a priori estimation of requisite
chemical parameters, such as solubilities and Henry's constant. Granted that considerable progress has
been made in process elucidation and modeling for chemical processes, such as photolysis and
hydrolysis, reliable estimates of the related fundamental physicochemical properties (i.e., rate and
equilibrium constants) have been achieved for only a limited number of molecular structures. The
values of these latter parameters, in most instances, must be derived from measurements or from the
expert judgment of specialists in that particular area of chemistry.
Mathematical models for predicting the transport and fate of pollutants in the environment
require reactivity parameter values—that is, the physical and chemical constants that govern reactivity.
Although empirical structure-activity relationships that allow estimation of some constants have been
available for many years, such relationships generally hold only within very limited families of
iv
-------
chemicals. On the other hand, we are developing computer programs that predict chemical reactivity
strictly from molecular structure for virtually all organic compounds. Our computer system called
SPARC (SPARC Performs Automated Reasoning in Chemistry) uses computational algorithms based
on fundamental chemical structure theory to estimate a large array of physical/chemical parameters. See
Table 1 for current SPARC physical property and chemical reactivity parameter estimation capabilities.
In every aspect of SPARC development, from choosing the programming environment to
building model algorithms or rule bases, system validation and verification were important criteria. The
basic mechanistic models in SPARC were designed and parameterized to be portable to any type of
chemistry or organic chemical structure. This extrapolatability impacts system validation and
verification in several ways. First, as the diversity of structures and the chemistry that is addressable
increases, so does the opportunity for error. More importantly, however, in verifying against the
theoretical knowledge of reactivity, specific situations can be chosen that offer specific challenges. This
is important when verifying or validating performance in areas where existing data are limited or where
additional data collection may be required. Finally, this expanded prediction capability allows one to
choose, for exhaustive validating, the reaction parameters for which large and reliable data sets do exist
to validate against. The SPARC models have been validated on more than 12,000 data points. The
verification and validation of the SPARC models will be presented in this report.
v
-------
Table 1. SPARC current physical and chemical properties estimation capabilities
Physical Property & Molecular
Descriptor
Molecular Weight
Polarizability
a, (3 H-bond
Microscopic bond dipole
Density
Volume
Refractive Index
Vapor Pressure
Viscosity
Boiling Point
Heat of Vaporization
Heat of formation
Diffusion Coefficient in Air
Diffusion Coefficient in Water
Activity Coefficient
Solubility
Gas/Liquid Partition
Gas/Solid Partition
Liquid/Liquid Partition
Liquid /Solid Partition
GC Retention Times
LC Retention Times
Status
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Mixed
Yes
Yes
UD
Mixed
Mixed
Yes
Yes
Yes
Mixed
Yes
Mixed
Yes
Mixed
Reaction
Conditions
Temp
Temp
Temp
Temp
Temp
Temp
Press
Temp
Temp
Temp, Press
Temp
Temp, Solv
Temp, Solv
Temp, Solv
Temp, Solv
Temp, Solv
Temp, Solv
Temp, Solv
Temp, Solv
Chemical Reactivity
lonization pKa in Water
lonization pKa in non-Aqueous Solution.
lonization pKa in Gas phase
Microscopic lonization pKa Constant
Zwitterionic Constant
Molecular Speciation
Isoelectric Point
Electron Affinity
Ester Carboxylic Hydrolysis Rate Constant
Hydration Constant
Tautomer Constant
E./2 Chemical Reduction Potential
Yes
Mixed
Mixed
Yes
Yes
Yes
Yes
Mixed
Yes
Mixed
Mixed
Mixed
Temp, pH
Temp, Solv
Temp
Temp, Solv, pH
Temp, Solv, pH
Temp, Solv, pH
Temp, Solv, pH
Temp , Solv
Temp , Solv
Temp, Solv, pH
Temp, Solv, pH
Mixed: Some capability exists but needs to be tested more, automated and/or extended.
Yes: Already tested and implemented in SPARC
UD: Under Development at this time
Press: Pressure, Temp: Temperature, Solv: Solvent
a: proton-donating site, P: proton-accepting site.
VI
-------
TABLE OF CONTENTS
INTRODUCTION 1
PREVIOUS PEER REVIEWS OF THE SPARC SYSTEM 2
ISSUES REGARDING VERIFICATION AND VALIDATION OF SPARC 4
SPARC COMPUTAIONAL APPROACH 6
SPARC PHYSICAL PROPERTIES MODELS 7
Validation of the SPARC Refractive Index Model 9
Validation of the SPARC Molecular Volume Model 11
Validation of the SPARC Vapor Pressure Model 12
Validation of the SPARC Boiling Point Model 14
Validation of the SPARC Activity Coefficient Model 15
Validation of the SPARC Solubility Model 17
Validation of the SPARC Mixed Solvents 18
Validation of the SPARC Partition Constants Models 19
Gas/liquid Partition Model 19
Liquid/Liquid Partition Model 20
Gas/Solid Partitioning Models 20
Liquid/Solid Partitioning Model 20
Validation of the SPARC Diffusion Coefficient in Air Model 22
vn
-------
SPARC CHEMICAL REACTIVITY MODELS 25
Validation of the SPARC lonization pKa in Water 25
Validation of the SPARC Carboxylic Acid Ester Hydrolysis Rate Constant 27
Validation of the SPARC Electron Affinity 29
SPARC MONOPOLE MODELS 30
Validation of the SPARC Monopole Models 30
QUALITY AS SURANCE 31
CONCLUSION 31
APPENDIX 32
REFERENCES 33
Vlll
-------
INTRODUCTION
Recent trends in environmental regulatory strategies dictate that EPA will rely heavily on
predictive modeling to carry out the increasingly complex array of exposure and risk
assessments necessary to develop scientifically defensible regulations. The pressing need for
multimedia, multistressor, multipathway assessments, from both the human and ecological
perspectives, over broad spatial and temporal scales, places a high priority on the development of
broad new modeling tools. However, as this modeling capability increases in complexity and
scale, so must the inputs. These new models will necessarily require huge arrays of input data,
and many of the required inputs are neither available nor easily measured. In response to this
need, researchers at NERL-Athens have developed the predictive modeling system SPARC
which calculates a large number of physical and chemical parameters from pollutant molecular
structure and basic information about the environment (media, temperature, pressure, pH, etc.).
Currently, SPARC calculates a wide array of physical properties and chemical reactivity
parameters for organic chemicals strictly from molecular structure. See Table 1.
SPARC has been in use in the Agency programs for several years, providing chemical
and physical properties to program offices (e.g., Office of Water, Office of Solid Waste and
Emergency Response, Office of Prevention, Pesticides and Toxic Substances) and Regional
Offices. Also, SPARC has been used in Agency modeling programs (e.g., the Multimedia,
Multi-pathway, Multi-receptor Risk Assessment (3MRA) and LENS3, a multi-component mass
balance model for application to oil spills) and to state agencies such as the Texas Natural
Resource Commission. The SPARC web-based calculators have been used by many employees of
various government agencies, academia and private chemical/pharmaceutical companies
1
-------
throughout the United States. The SPARC web version performs approximately 50,000-100,000
calculations each month. (See the summary of usage of the SPARC web version in the Appendix).
PREVIOUS PEER REVIEWS OF THE SPARC SYSTEM
Over the lifespan of its development, the SPARC computer system has undergone
numerous (and various types of) reviews that have helped to establish its validity. For example, we
have published 10 journal articles on the SPARC computer system, each of which has undergone
extensive peer-review (See references 2-11). Also, the SPARC computer system underwent an
EPA Science Advisory Board (SAB) review in 1991, which was relatively early in its development
stage. This multi-day review gave the SPARC development team the opportunity to demonstrate
the system, and to discuss its modeling philosophy with experts in environmental science. Their
comments on the system were very favorable, and they provided important input on further system
development. Following is a brief excerpt from the SAB's written report. "For a program still in
development, progress is excellent. Resources should be made available to complete the
documentation and conduct extensive testing of the model" [12].
Also, the SPARC computer system has been included, along with other projects at our
Division, in several major peer reviews, once in 1997 and again in 2000. These reviews were
conducted by "blue-ribbon" panels of scientists from outside of EPA. Again, the comments on
SPARC were laudatory and they provided important input to model development. For example,
the following is an excerpt from the 1997 peer review panel's report on the SPARC project. "The
review panel is extremely impressed with the quality and productivity of this broad project, as well
as with the presentation, the speaker, and the body of work that he summarized in such a
remarkable fashion. This effort represents a central cog in the entire ERD program in
environmental chemistry, as well as a key component of the Division's programmatic support to
the broader agency. Moreover, this body of work represents a truly impressive service to the
-------
larger environmental chemistry community. EPA must find ways of providing permanent, long-
term support for the commitment to this effort. It should also work toward making this service fully
available via the internet. This project truly represents a key component of the NERL/Athens
scientific endeavor "[13].
Furthermore, the SPARC developers have frequently engaged in informal consultation with
leaders in relevant fields of science throughout the SPARC model design. These scientists include
the late Dr. Robert Taft of the University of California, Dr. John Garst and Dr. Bruce King of the
University of Georgia, Dr. Ralph Dougherty of the Florida State University and Dr. Samuel
Yalkowski of the University of Arizona.
In summary, the SPARC system has now been extensively reviewed by many renowned
scientists outside of EPA and in many different peer-review processes. Reviewer comments have
always been favorable and the suggestions of these scientists have always been used to improve
further model development. This type of "open communication" with leaders in various fields of
science improves and helps to establish the validity of the SPARC models.
Although development of the SPARC program has been aimed at use in environmental
assessments, these physicochemical models have widespread applicability in the academic and
industrial communities For example, the SPARC program has been used at several universities as
an instructional tool to demonstrate the applicability of physical organic models to the quantitative
calculation of physicochemical properties (e.g., graduate class taught by the late Dr. Robert Taft at
the University of California). Also, the SPARC calculator has great potential for aiding industry
(such as Pfizer, Merck, Pharmacia & Upjohn, etc.) in the areas of chemical manufacturing and
pharmaceutical and pesticide design.
-------
ISSUES REGARDING VERIFICATION AND VALIDATION OF SPARC
To adequately convey the importance of verification and validation of SPARC models, it
is necessary to first describe briefly, and in general terms, the SPARC modeling approach and
philosophy. Indeed, it is useful to compare and contrast the SPARC approach to that of more
conventional models for predicting physicochemical parameters.
Most models that predict a given physicochemical property (e.g., solubility, boiling point,
etc.) are based, in a very direct way, on experimental data for that property for a limited training
set of chemicals. Model development involves finding the best correlations between various
descriptors of chemical structure and the observed property values. These descriptors are
subsequently used to construct a model that adequately "recalculates" the training (or
calibration) data set. Then, to validate, one must demonstrate that the empirical model also
accurately predicts property values for chemicals not included in the training set, but whose
experimental values are known. These data are often called the validation set. In order to
predict a new physicochemical property (e.g., octanol/water partition coefficient), the entire
process must be repeated, requiring new training and validation data sets for each new property.
On the other hand, with SPARC, experimental data for physicochemical properties (such
as boiling point) are not used to develop (or directly impact) the model that calculates that
particular property. Instead, physicochemical properties are predicted using a few models that
quantify the underlying phenomena that drive all types of chemical behavior (e.g., resonance,
electrostatic, induction, dispersion, H-bonding interactions, etc.). These mechanistic models
were parameterized using a very limited set of experimental data, but not data for the end-use
properties that will subsequently be predicted. After verification, the mechanistic models were
-------
used in (or ported to) the various software modules that calculate the various end-use properties
(such as boiling point). It is critical to recognize that the same mechanistic model (e.g., H-
bonding model) will appear in all of the software modules that predict the various end-use
properties (e.g., boiling point) for which that phenomenon is important. Thus, any comparison
of SPARC-calculated physicochemical properties to an adequate experimental data set is a true
model validation test — there is no training (or calibration) data set in the traditional sense for
that particular property. The results of validation tests on the various SPARC property models
are presented below in the sections devoted to each property.
The unique approach to SPARC modeling also impacts our strategy for module
verification. For example, when a mechanistic model is updated or improved by incorporating
new knowledge, the impact on all of the various end-use parameters must be assessed. Toward
this end, we have developed quality assurance software that executes each quarter. This
software runs the various property modules for a large number of chemicals (4200 data-point
calculations) and compares the output to historical results obtained over the life-span of the
SPARC program. (Note that, early in our developmental stage, output of all SPARC modules
were compared to hand calculations with selected chemicals to the extent possible. Satisfactory
results were obtained prior to proceeding with further development). In this way, we ensure that
existing parameter models still work correctly after new capabilities and improvements are
added to SPARC. This also ensures that the computer code for all property and mechanistic
models are fully operational. Since the same approach to verification was taken for all property
modules, and since they are all driven by the same verified mechanistic models, we will not
discuss verification in the following sections devoted to each property.
-------
SPARC COMPUTAIONAL APPROACH
SPARC does not do "first principles" computation; rather, it analyzes chemical structure
relative to a specific reactivity query much as an expert chemist might. SPARC utilizes directly the
extensive knowledge base of organic chemistry. Organic chemists have established the types of
structural groups or atomic arrays that impact certain types of reactivity and have described, in
"mechanistic" terms, the effects on reactivity of other structural constituents appended to the site of
reaction. To encode this knowledge base, a classification scheme was developed that defines the
role of structural constituents in affecting or modifying reactivity. Furthermore, models have been
developed that quantify the various "mechanistic" descriptions commonly utilized in structure-
activity analysis, such as induction, resonance and field effects. SPARC execution involves the
classification of molecular structure (relative to a particular reactivity of interest) and the selection
and execution of appropriate "mechanistic" models to quantify reactivity. In brief, the SPARC
model consists of a set of core models describing intra/intermolecular interactions that are linked by
the appropriate thermodynamic relationships to provide estimates of reactivity parameters under
desired conditions such as temperature, pressure and pH. The details of SPARC computational
methods are presented in a companion U.S. E.P.A report, "Prediction of Chemical Reactivity
Parameters and Physical Properties of Organic Compounds from Molecular Structure Using
SPARC" [14]. Hence, only an overview will be given here.
For physical properties, intermolecular interactions are expressed as a summation over all the
interaction forces between molecules (i.e., dispersion, induction, dipole and H-bonding). Each of
these interaction energies is expressed in terms of a limited set of molecular-level descriptors
(volume, molecular polarizability, molecular dipole, and H-bonding parameters) that, in turn, are
-------
calculated from molecular structure. For chemical reactivity, molecular structure is broken into
functional units. Reaction centers with known intrinsic reactivity are identified and the impact on
reactivity of appended molecular structure is quantified using mechanistic perturbation models.
A "toolbox" of mechanistic perturbation models has been developed that can be
implemented where needed for a specific reactivity query. Resonance models were developed and
validated on more than 5000 light absorption spectra [1, 2], whereas electrostatic interaction
models were developed and validated on more than 4500 ionization pKas in water [3-8]. Solvation
models (i.e., dispersion, induction, H-bond and dipole interactions) have been developed and
validated on more than 8000 physical property data points on properties such as vapor pressure,
boiling point, solubility, Henry's constant, GC chromatographic retention times, Kow, etc [3, 9,
10]. The SPARC computational approach is based on blending well known, established methods
such as SAR (Structure Activity Relationships) [15, 16], LFER (Linear Free Energy Relationships)
[17, 18] and PMO (Perturbed Molecular Orbital) theory [19, 20]. SPARC uses SAR for structure
activity analysis, such as induction, resonance and field effects. LFER is used to estimate
thermodynamic or thermal properties and PMO theory is used to describe quantum effects such as
charge distribution delocalization energy and polarizability of the TT electron network.
SPARC PHYSICAL PROPERTIES MODELS
For all physical properties (e.g., vapor pressure, boiling point, activity coefficient, solubility,
partition coefficients, GC/LC chromatographic retention times, diffusion coefficients, etc.), SPARC
uses one master equation to calculate characteristic process parameters:
ACjprocess ~~ A (_T Interaction A (_T other ^ '
-------
where AGinteraction describes the change in the energy associated with the intermolecular interactions
accompanying the process in question. For example, in liquid to gas vaporization, AGinteraction
describes the difference in the energy associated with intermolecular interactions in the gaseous
phase versus that associated with interactions in the liquid phase. The intermolecular interaction
forces between the molecules are assumed to be additive. The AGother lumps all non-interaction
energy components such as entropy changes associated with mixing or expansion, and changes in
internal molecular (vibrational, rotational) energies. At the present time, the intermolecular
interactions in the liquid phase are modeled explicitly, interactions in the gas phase are ignored, and
molecular interactions in the crystalline phase are extrapolated from the subcooled liquid state using
the melting point. The 'non-interaction' entropy components are process specific and will be
described later, in the vapor pressure and the activity coefficient models. The intermolecular
interactions in the liquid phase are expressed as a summation over all the mechanistic components:
'-^Interaction = ^^ Dispersion + ^^ Induction + ^^ Dipole-dipole + '~^JH-Bond (2)
Each of these interaction mechanisms is expressed in terms of a limited set of pure
component descriptors (liquid density-based volume, molecular polarizability, microscopic bond
dipole, and hydrogen bonding parameters), which in turn are calculated strictly from molecular
structure [3, 9].
Dispersion interactions are present in all molecules, including polar and non-polar
molecules. Induction interactions are present between two molecules when at least one of them has a
local dipole moment. Dipole-dipole interactions exist when both molecules have local dipole
moments. H-bonding interactions exist when a; |3j or a,j (3; products are non zero, where a
-------
represents the proton donation strength and |3 represents the proton acceptor strength. In SPARC, all
the physical property estimations derive from a common set of core models describing
intra/intermolecular interactions, and require as user inputs molecular structure (both solute and
solvent(s)) and reaction conditions of interest (temperature, pressure, etc.). The self-term, AG;
(solute-solute) interaction model is used to describe the vapor pressure at 25° C. The self terms, AG;;
and AGjj (solvent-solvent) plus the cross term, AGy (solute-solvent) interactions, are required to
describe the solute, i, activity coefficient in any solvent, j at 25° C.
Like the chemical reactivity models, the AG;;, AGy and AGjj models have been extended and
validated on numerous physical properties under different reaction conditions such as temperature,
pressure and solvents. The self-term interaction model has been tested on a large number of vapor
pressures, boiling points, diffusion coefficients and heat of vaporization. Likewise, the solute-
solvent interaction model has been validated on activity coefficients, solubilities, partition
coefficients and GC/LC chromatographic retention times in any solvent at any temperature.
Validation of the SPARC Refractive Index Model
The molecular polarizability and volume can be related to the index of refraction (n) using
the Lorentz-Lorenz equation. For our units of cmVmole for volume (V) and A3/molecule for
polarizability (P), the Lorentz-Lorenz equation can be written as
-------
n2 -1 _ 4;r(0.6023,P)
^72 " W (3)
The refractive index output was initially verified by comparing the SPARC prediction
value to hand calculations for selected key molecules. The refractive index calculator was trained
on 325 non-polar and polar organic compounds at 25° C then validated on 578 organic
compounds at 25° C [9, 10] as shown in Figure 1. The statistical performance for the SPARC
refractive index calculator is shown in Table 2. See reference 9 for sample hand calculations.
10
-------
Table 2. SPARC Physical and Chemical Properties Calculator Statistical Performance
versus Observations
Property
Refractive Index
Volume
Vapor Pressure
Boiling Point
Heat
of Vaporization3
Diffusion
Coefficient in Air4
Activity Coefficient
Solubility
Distribution
Coefficient
Henry's Constant
GC Retention Time2
LC Retention Time
Gas pKa3
Non-aqueous pKa3
pKa in water
Electron Affinity
Ester Carboxylic
Hydrolysis Rate
Tautomer Constant3
Hydration Constant3
E./2 Chemical
Reduction3
Units
N/A
g/cm3
log arm
°C
Kcal/mole
cm2/s
log MF5
logMF
N/A
M/L6
Kovtas
Kovtas
Kcal
Kcal
Kcal/1.36
e.V.
M-V1
Kcal/1.36
Kcal/1.36
e.V
Total #
Molecule
578
1440
747
4000
1263
108
491
647
623
286
271
295
125
400
300
4338
260
1470
36
27
352
RMS
0.007
1.97
0.15
5.71
0.301
0.003
0.064
0.40
0.43
0.34
0.10
10
0.095
2.25
1.90
0.356
0.14
0.37
0.3
0.43
0.18
R2
0.997
0.999
0.994
0.999
0.993
0.994
0.998
0.987
0.983
0.990
0.997
0.998
0.992
0.999
0.960
0.994
0.98
0.968
0.950
0.744
0.95
Reaction Conditions
Temp/Solvent
25
25
25
0.1-1520 torr
25, Boiling Point
25
25, 41 solvents
25, 21 solvents
25 Octanol, Toluene CC14,
Benzene,
Cyclohexane, Ethyl Ether
25, Water
25, Hexadecane
25-190, Squalane,B18
25, Water/Methanol
25 , Alcohols, Aceteonitrile,
Acetic acid, DMF1, THF1,
pyridine
25-100, Water
Gas
25-130, Water, Acetone,
Alcohols, Dioxane,
Aceteonitrile
25, Water
25, Water, Alcohols, DMF1
Aceteonitrile, DMSO1
1 DMF: N,N -dimethylforamide
DMSO: Dimethyl sulfoxide
HF: Tetrahydrofuran
2. GC retention times in SE-30, OV-101 and PEG-20M liquid stationary liquid phase is
not included in this report.
3 See the companion SPARC report [14]
4. Models were developed after the HWIR exercise.
5. MF: mole fraction
6. M/L unit is (mole/L)/(mole/L); unitless
11
-------
1
jj
ro
O
o
a:
1.52
1.47
1 .4 2
1.37
1.32
y = 1 . 0 1 0 1 x - 0.0152
O bserved
1.32
1.52
Figure 1. SPARC-calculated versus observed refractive index at 25° C. The RMS (Root Mean
Square) deviation was 0.007 and R2 was 0.997.
Validation of the SPARC Molecular Volume Models
The zero order density-based molecular volume at 25° C is expressed as
rag-4) (4)
where V;frag is the volume of the ith molecular fragment and A; is a correction to that volume based
on both the number and size of fragments attached to it. The V;fragare determined empirically from
measured liquid-density based volumes, and then stored in the SPARC database. The zero order
volume at 25° C is further adjusted for changes resulting from dipole-dipole and hydrogen bonding
intermolecular interactions:
V, = V2°5 + Adlpole _
dipole - dipole
A
H - bond
(5)
12
-------
where D; is the weighted sum of the local dipole for the molecule, and a and P are the H-bonding
parameters of potential proton donor and proton acceptor sites within the molecule, respectively.
Adipoie-dipoie and AH-bond are adjustment constants due to dipole-dipole and H-bonding, respectively.
The final molecular volume at any temperature T is then expressed as a polynomial expansion in
(T-25) corrected for H- bonding, dipole density and polarizability density interactions [9, 14].
The molecular volume can be calculated within 2 cm3 mole"1 for most organic molecules.
Figure 2 shows the SPARC-calculated versus observed molecular volumes for both polar and non-
polar compounds at 25° C. The statistical performance for the volume calculator is in Table 2. See
reference 9 for sample hand calculations.
o
E
ra
ra
O
6
Q.
V)
= 0.9966X + 0.2982
200
400
600
800
1000
Observed (cm /mole)
Figure 2. SPARC-calculated vs. observed-liquid density based volume at 25° C for 1440 organic
molecules. The RMS deviation was 1.97 cm3 mole"1 and R2 was 0.999.
Validation of the SPARC Vapor Pressure Model
The saturated vapor pressure is one of the most important physiochemical properties of pure
compounds. By 1978, vapor pressure data (as a function of temperature) were available for more
13
-------
than 7000 organic compounds [21]. Despite the frequency of reporting in the published literature,
the number of compounds where the vapor pressure was truly measured and not extrapolated to 25°
C from higher temperature measurements, is limited. Most of the measured 25° C vapor pressure
data are for compounds that are either pure hydrocarbons or molecules that have relatively small
dipole moments and/or weak hydrogen bonds. There is a pressing need to predict the vapor pressure
of those compounds that have not been measured experimentally. In addition to being highly
significant in evaluating a compound's environmental fate, the vapor pressure at 25° C provides an
excellent arena for developing and testing the SPARC self interaction physical process models.
The vapor pressure, vp°; of a pure solute, i, can be expressed as function of all the
intermolecular interaction mechanisms, A G;; (interaction), as
o - AG. (Interaction) , .
log vpi = — + LogT + C (6)
g Pl 2.303RT
where log (T) + C describes the change in the entropy contribution associated with the volume
change in going from the liquid to the gas phase. The crystal energy term (given in reference 14),
CE, must be added to equation 6 for molecules that are solids at 25° C, the CE contribution becomes
important, especially for rigid structures such as aromatic or ethylenic molecules that have high
melting points [14].
The vapor pressure computational algorithm output was initially verified by comparing the
SPARC prediction of the vapor pressure at 25° C to hand calculations for key molecules. Since the
SPARC self interactions model, AGu, was developed initially on this property, the vapor pressure
model undergoes the most frequent validation tests. The calculator was trained on 315 non-polar
14
-------
and polar organic compounds at 25° C. Figure 3 presents the SPARC-calculated vapor pressure at
25° C versus measured values for 747 compounds. The SPARC self-interactions model can predict
the vapor pressure at 25° C within experimental error over a wide range of molecular structures and
measurements (over 8 log units). For simple structures, SPARC can calculate the vapor pressure to
better than a factor of 2. For complex structures such as some of the pesticides and pharmaceutical
drugs where dipole-dipole and/or hydrogen bond interactions are strong, SPARC calculates the
vapor pressure within a factor of 3-4. The statistical performance for the vapor pressure calculator
is shown in Table 2. See references 9 and 14 for sample hand calculations. The vapor pressure
model was also tested on the boiling point and heats of vaporization [9, 14].
+j
re
O)
_re
3
"re
9
o
OL
Q.
(A
y=0.9942x-0.0117
1
- 1
-3
-5
-7
-7
-5 -3 - 1
Observed (log atm)
Figure 3. SPARC-calculated vs. observed log vapor pressure for 747 organic molecules at 25° C.
The figure includes all the vapor pressure measurements (real not extrapolated) we found in the
literature. The RMS deviation error was 0.15 log atm and R2 was 0.994.
Validation of the SPARC Boiling Point Model
SPARC estimates the boiling point for any molecular species by varying the temperature
at which a vapor pressure calculation is done. When the vapor pressure equals the desired
15
-------
pressure, then that temperature is the boiling point at that pressure. The normal boiling point is
calculated by setting the desired pressure to 760 torr. Boiling points at a reduced pressure can be
calculated by setting the desired pressure to a different value.
SPARC temperature dependence models were developed initially on the boiling point. The
boiling point calculator was trained on 1900 boiling points for a wide range of non-polar and polar
organic compounds. The calculator was validated against 4000 boiling points measured at different
pressures ranging from 0.05 to 1520 torr spanning a range of over 800° C as shown in Figure 4.
700 ~r
y=0.9981x-0.1994
o
D
.o
ns
s>
o
0.
V)
200
-300
-300
-100
700
O bserved (° C )
Figure 4. SPARC-calculation versus observed 4000 boiling points for pressure ranging from 0.1 to at
1520 torr. The Total RMS deviation was 5.71° C. The RMS deviation for polar molecules was 8.2°
C and R2 was 0.9988, while for non-polar molecules the RMS was 2.6° C and R2 was 0.9995.
Validation of the SPARC Activity Coefficient Model
For a solute, i, in a liquid phase,], at infinite dilution, SPARC expresses the activity coefficient as
(7)
-RT log y* = ZAGy (Interaction) + RT (log— + —
Vj 2.juj
16
-------
where V; and Vj are the molecular volume of the solute and the solvent, respectively. The last term
is the Flory-Huggins excess-entropy-of-mixing contribution in the liquid phase for placing a solute
molecule in the solvent [3, 14].
The activity coefficient computational algorithm output was initially verified by comparing
the SPARC prediction to hand calculations for key molecules. The SPARC activity coefficient
calculator was trained on 211 activities for a wide range of organic molecules. Figure 5 presents the
validation for SPARC-calculated log activity coefficients versus measured values for 491
compounds at 25° C in 41 different solvents. The SPARC activity coefficient test statistical
parameters are shown in Table 2. The activity coefficients calculator was also tested on the
solubility in more than 20 different solvents and partition coefficients in more than 18 different
solvents. See following sections for more details.
o
ra
o
E
o>
o
_o
ra
O
8
7
6
5
4
3
2
1
0
-1
-2
y = 0.9983X - 0.0001
-2
0246
Observed (log mole fraction)
Figure 5. SPARC-calculated versus observed log activity coefficients at infinite dilution for 491
compounds in 41 solvents including water. Only 15% of these compounds have strong dipole-
dipole and/or H-bond interactions. The RMS deviation was 0.064 log mole fraction and with an
R2 of 0.998.
17
-------
Validation of the SPARC Solubility Model
SPARC does not calculate the solubility from first principles, but rather from the infinite
dilution activity coefficient model discussed previously. SPARC first calculates the infinite dilution
activity coefficient, y°°; when log y°° is greater than 2, the mole fraction solubility can be reliably
estimated as xsd = 1/Y°°- However, when the log y°° is calculated to be less than 2, this approximation
fails. In these cases, y°° is greater than yso1 and SPARC would underestimate the solubility using the
inverse relationship. In order to overcome this limitation, SPARC employs an iterative calculation.
SPARC sets the initial guess of the solubility as K^gaess = l/y°°- SPARC then 'prepares' a mixed
solvent that is xsolguess in the solute and (1- xsolguess) in the solvent. SPARC then recalculates y°° in the
'new' solvent and the corresponding xsolguess- This process is continued until y°° converges to 1
(miscible). The solubility calculator spans more than 12 log mole fraction as shown in Figure 6.
The RMS deviation was 0.40 log mole fraction, which was close to the experimental error.
SPARC estimates the solubility for simple organic molecules to better than a factor of 2 (0.3 log
mole fraction) and within a factor of 4 (0.6 log mole fraction) for complicated molecules like
pesticides and pharmaceutical drugs. The RMS deviation for the solids compounds is 3 times
greater than the RMS deviation for liquids compounds due to the crystal energy contributions. For
more details see reference 14. The statistical parameters for calculated log solubility for 647
organic molecules in 21 different solvents including water at 25° C are shown in Table 2.
18
-------
I
3
-1 5
y = 0.9944X - 0.043 1
-1 5
-1 0 -5
Observed (log mole fraction)
Figure 6. Test results for SPARC calculated log solubilites for 260 compounds. The RMS
deviation is 0.321 and R2 is 0.991. The RMS deviation for 119 liquid soluibilties is 0.135 and R2
is 0.997 while for the 141 solids compounds the RMS deviation is 0.419 and R2 is 0.985.
Validation of the SPARC Mixed Solvents Model
SPARC can handle solvent mixtures for a large number of components. However, speed
and memory requirements usually limit the number of solvent components to less than twenty on
a PC. The user specifies the name and volume fraction for each solvent component. Each of the
solvent components must have been previously initialized as a solvent. SPARC will allow the
user to specify a name for the mixture so that it can be used later as a 'known' solvent. The
activity coefficients (or solubility) of molecules in binary solvent mixtures have been tested and
appear to work well. Figure 7 shows the calculated log y in a water/methanol mixture versus
measured values. For more details see reference 14.
19
-------
- 6 n
y = 0.9535X + 0.1422
« Q
0
1234
Observed (log mole fraction)
Figure 7. SPARC-calculated versus observed log activities for 120 compounds in water/methanol
mixed solvent at 25° C. The RMS deviation error was 0.18 and the R2 was 0.980.
Validation of the SPARC Partition Constants Models
All partition (Liquid/Liquid, Liquid/Solid, Gas/Liquid, Gas/Solid) constants are
determined by calculating the activity coefficient of the molecular species of concern in each of
the phases without modification or extra parameterization to the activity coefficient model.
Gas/liquid (Henry's Constant) Model
Henry's constant may be expressed as
(8)
where vp;° is the vapor pressure of pure solute i (liquid or subcooled liquid) and yy00 is the activity
coefficient of solute (i) in the liquid phase (j) at infinite dilution. SPARC vapor pressure and activity
coefficient models can be used to calculate the Henry's constant for any solute out of a
20
-------
mixed solute-solvent liquid phase. An application of Henry's law constant for the prediction of gas-
liquid chromatography retention time is given in the companion SPARC report [14].
Liquid/Liquid Partitioning Model
SPARC calculates the liquid/liquid partition constant, such as the octanol/water distribution
coefficient, by simply calculating the activity of the molecular species in each of the liquid phases as
logKuqi/iw = lo§7, -lo§7, , + \°gRm (9)
^ ^ / hq2 I liql
where the y^s are the infinite dilution activities in the two phases and Rm is the ratio of the
molecularites of the two phases (Mi/M2). Although octanol/water partition coefficients are widely
used and measured, the SPARC system does not limit itself to this calculation. SPARC can
calculate the liquid/liquid partition coefficient for any two immiscible phases.
Gas/Solid Partitioning Model
SPARC calculates gas/solid partitioning in a manner similar to gas/liquid partitioning. For
the solid phase, the solvent self-self interactions, AGy, are dropped from the calculation when one of
the phases is solid. This type of modeling will be useful for calculating retention times for capillary
column gas chromatography.
Liquid/Solid Partitioning Model
SPARC calculates liquid/solid partitioning in a manner similar to liquid/liquid partitioning.
For the solid phase, the solvent self-self interactions, AGy, are dropped from the calculation.
21
-------
The gas/liquid models have been extensively tested against observed Henry's constant
measurements. The two largest data sets are air/water and air/hexadecane systems. The liquid/solid
and gas/solid partitioning models are implemented in code but have not been extensively tested. The
liquid/liquid partitioning models are the most extensively tested partitioning models due to the large
octanol/water data sets available. The statistical parameters for SPARC-calculated partition
constants in many solvents at 25° C are shown in Table 2. Figure 8 shows calculated versus
observed Henry's constant for compounds dissolved in hexadecane. Figure 9 shows the current
general performance of SPARC for log Ksoivent/water, where the solvents are carbon tetrachloride,
benzene, cyclohexane, ethyl ether, octanol and toluene. Figure 10 displays a comparison of the EPA
Office of water (OW) recommended observed octanol-water distribution coefficients versus SPARC
and C log P calculated values. The RMS deviation and R2 values were is 0.18 and 0.996
respectively for SPARC and 0.44 and 0.978 respectively for ClogP calculated values [22].
TJ
ra
O
4 -,
2 -
0 -
-2 -
-4 -
-6 -
-8 -
-1 0
y = 0.998x - 0.006
-1 o
-8 -6 -4 -2 0
O bserved (m ole/L)/m ole/L
Figure 8. Observed vs. SPARC-calculated Henry's constants for 271 organic compounds in
hexadecane. The RMS deviation was 0.1, while the R2 was 0.997.
22
-------
o
10
8
6
y = 0.9721x +0.0363
-8
-3
7
Observed log Kow
Figure 9. SP ARC-calculated versus observed log distribution coefficients Ksoivent/waterfor 623
organic compounds in six solvents at 25° C. The RMS deviation was 0.38 and R2 was 0.983.
Q.
D)
£
O
C
re
0)
•+J
JO
D
-------
a wide variety of molecules. This equation is that of Wilke and Lee [23], which for binary diffusion
coefficient is expressed as:
DAB = [3.03 - (0.98 / M^l)]( Iff3)-
r
(10)
where DAB is the binary diffusion coefficient in cm /s, T is the temperature in K, MA and MB are
-i-i
the molecular weights of A and B in g/mol and MAB is 2[(1/MA) + (I/Ms)]" and P is the pressure
in bar. The QD is a complex function of T and has been accurately determined by Neufeld [24].
SPARC predicts gas phase binary diffusion coefficients at any temperature and pressure
to better than 6% as shown in Figure 11. The statistical parameters are in Table shown 2.
_o
«8
9
o
0 .2
0.15
0 . 1
0.05
y = 0.9762X + 0.0021
0.05
0 . 1
0.15
0 .2
Observed (cm Is)
Figure 11. SPARC-calculated vs. observed diffusion coefficient. The RMS deviation was 0.003.
The overall SPARC physical properties training set output is shown in Figure 12. The
training set includes vapor pressure (as a function of temperature), boiling point (as a function of
pressure), diffusion coefficients (as a function of pressure and temperature), heat of vaporization (as
function of temperature), activity coefficient (as a function of solvent), solubility (as a function of
24
-------
solvent and temperature), GC retention times (as a function of stationary liquid phase and
temperature) and partition coefficients (as a function of solvent). This set includes more than 50
different pure solvents (see Table 3) as well as 18 mixed solvent systems. The observed measured
values for the training and validations sets were from many sources such as references 26-34.
For other SPARC physical properties models such as GC/LC retention time in polar and
non-polar liquid phase, heat of vaporization and diffusion coefficient in water, see reference 14.
_0
TO
o
300
1 00
-100
-300
y = 0.9981X + 0.0878
-300
-100 100
O bserved
300
Figure 12. SPARC-calculated vs. 2400 observed training set physical property values. The aggregate
RMS is 0.29 and R2 is 0.997. For more details see text.
Table 3. Solvents that have been tested in SPARC
Chloroform
1-propanol
isobutanol
benzyl ether
cyclohexane
cyanohexane
heptane
methanol
nitroethane
nonanenitrile
quinoline
1-butanol
butanone
acetone
benzene
decane
ethanol
hexane
nonane
octane
squalene
phenol
1-chloro hexadecane
1-nitro propane
2-nitro propane
benzylchloride
bromobenzene
dioctyl ether
hexadecane
1 -butyl chloride
nitro cyclohexane
pentadecane nitrile
1,2,4 trichlorobenzene
1-dodecanol OV-101
2-dodecanone isopropanol
aceteonitrile PEG-20M
benzonitrile SE-30
butronitrile pyridine
cyano cyclohexane water
heptadecane squalane
nitrobenzene 1-me naphthalene
nitro methane 2-me naphthalene
isoquinoline m-cresol
hexafluorobenzene p-xylene
25
-------
SPARC CHEMICAL REACTIVITY MODELS
SPARC reactivity models have been designed and parameterized to be portable to any
chemical reactivity property and any chemical structure. For example, chemical reactivity models
are used to estimate ionization pKa, zwitterionic constant, isoelectric point and speciation
fractions as a function of pH. The same reactivity models are used to estimate gas phase electron
affinity and ester hydrolysis rate constants in water and in non-aqueous solutions.
Validation of the SPARC pKa in water Models
Like all chemical reactivity parameters addressed in SPARC, molecular structures are
broken into functional units called the reaction center and the perturber in order to estimate pKa
in water. The reaction center, C, is the smallest subunit that has the potential to ionize and lose a
proton to a solvent. The perturber, P, is the molecular structure appended to the reaction center,
C. The pKa of the reaction center is adjusted for the molecule in question using the mechanistic
perturbation models. The pKa for a molecule of interest is expressed in terms of the
contributions of both P and C.
pKa = (pKa)c + 5p(pKa)c (11)
where (pKa)c describes the ionization behavior of the reaction center, and 5p(pKa)c is the change
in ionization behavior brought about by the perturber structure given as
8p(pKa)c = 5elepKa + 5respKa + 5solpKa+... (12)
where 5respKa, 5eiepKa and 5soipKa describe the differential resonance, electrostatic and solvation
effects of P on the initial and final states of C, respectively.
26
-------
The SPARC pKa calculator was trained on 2500 organic molecules, then validated on
4338 pKa's (4550 including carbon acid) in water as shown in Figure 13 and Table 4. The
calculator was tested for multiple ionization's up to the 6th (simple organic molecules) and 8th (azo
dyes) for molecules with multiple ionization sites. In addition, the pKa models were tested on all
the literature values we found for zwitterionic constants (12 data points), the thermodynamic
microscopic ionization constants, pk;, of molecules with multiple ionization sites (120
measurement data points, the RMS deviation error is 0.5), the corresponding complex speciation
as a function of pH and the isoelectric points (29 measurement data points) in water. The
diversity and complexity of the molecules used was varied over a wide range in order to develop
more robust models during the last few years. Hence, the SPARC pKa models are now very
robust and highly tested against almost all the available experimental literature data.
While it is difficult to give a precise standard deviation of a SPARC calculated value for
any given individual molecule, in general SPARC can calculate the pKa for simple molecules
such as oxy acids and aliphatic bases and acids within ±0.25 pKa units; ±0.36 pKa units for most
other organic structures such as amines and acids; and ±0.41 pKa units for =N and in-ring N
reaction centers and for complicated structures. Where a molecule has more than six ionization
sites (n > 6), the expected SPARC error is ±0.65 pKa units. For more details see reference 14.
Table 4. Statistical Parameters of SPARC pKa Calculations
Set
Simple organic compounds
Azo dyes compounds
IUPAC compounds1
Training
793
50
2500
R2
0.995
0.991
0.994
RMS
0.235
0.550
0.356
Test
2000
273
43382
R2
0.995
0.990
0.994
RMS
0.274
0.630
0.370
1. Observed values are from many ref such as 35-36
2. Carbon acid pKas are not included
27
-------
1 8 -
o
oi
-2 -
-1 2
y = 0.9925X + 0.0189
-1 2
•2 8
Observed pKa
1 8
FigurelS. SPARC-calculated versus observed for 4338 pKa's of 3685 organic compounds. The
RMS deviation was equal to 0.37. This test does not include carbon acid reaction center. The
majority of the molecules are complex compounds. Some of the molecules such as azo dyes
have 8 different ionization sites.
Validation of the SPARC Carboxylic Acid Ester Hydrolysis Rate Constant Models
Reaction kinetics were quantitatively modeled within the chemical equilibrium
framework described previously for ionization pKa in water. It was assumed that a reaction rate
constant could be described in terms of a pseudo equilibrium constant between the reactant and
transition states. SPARC therefore follows the modeling approach described for pKa. For these
chemicals, reaction centers with known intrinsic reactivity are identified and the reaction rate
constants expressed by perturbation theory as
where log k is the log of the rate constant of interest; log kc is the log of the intrinsic rate constant
of the reaction center and Aplog kc denotes the perturbation of the log rate constant due to the
appended structure.
28
-------
The ester hydrolysis rate constant models have been tested to the maximum extent possible
as function of temperature and solvent. The RMS deviation error for 1470 hydrolysis rate constants
in 6 solvents and at different temperature was 0.37 as shown in Figure 14. In this test, there were
653, 667 and 150 base, acid and general base catalyzed calculations performed as shown in Table 5
[14, 25].
Table 5. Statistical Parameters of SPARC Calculated Hydrolysis Rate Constants (M'V)
Solvent
Water
Acetone/Water
Ethanol/Water
Methanol/Water
Dioxnae/Water
Aceteonitrile/Water
Total Molecules
Base
No RMS R2
142
143
105
150
90
24
654
0.39 0.98
0.34 0.83
0.29 0.83
0.36 0.78
0.47 0.75
0.3 0.97
0.37 0.96
Acid
No RMS R2
383
208
39
22
15
N/A
667
0.36 0.98
0.33 0.96
0.17 0.98
0.22 0 .95
0.16 0.87
0.37 0.97
Gbase
No RMS R2
51
73
9
N/A
17
N/A
150
0.34 0.98
0.36 0.96
0.1 0.99
0.47 0.67
0.39 0.97
The observed-measured values are from many references such as 37-40
-a
-S2
03
-§ o
CO
O
O
Q_
CO
-6
-12
y = O.9744X - O.O892
-12
-6 O
Observed (M~V1)
6
Figure 14. SPARC-calculated versus observed hydrolysis rate constants for base, acid and
general base in six different solvents and at different temperatures. The aggregate RMS was 0.37.
29
-------
Validation of the SPARC Electron Affinity (EA) Models
As was the case for pKa, the SPARC computational procedure starts by locating the
potential sites within the molecule at which a particular reaction of interest could occur. In the
case of EA these reaction centers, C, are the smallest subunit(s) that could form a molecular
negative ion. Any molecular structure appended to C is viewed as a "perturber" (P). EA as
expressed in terms of the summation of the contributions of all the components, perturber(s) and
reaction center(s), in the molecule:
EA = [(EA)c+Sp(*EA)J (14)
c=l
where the summation is over n, which is defined as the number of reaction centers in the
molecule. (EA)C is the electron affinity for the reaction center. 5P(AEA)C is a differential quantity
that describes the change in the electron affinity behavior affected by the perturber structure.
In the estimation of EA, there was no modifications to any of the pKa models or any extra
parameterization for P to calculate electron affinity from ionization pKa models other than inferring
the electronegativity and the electron affinity susceptibility of the reaction centers (C) to
electrostatic and resonance effects [4].
The EA models have been tested to the maximum extent possible on all the gas phase
electron affinity measurements reported by Kebarle, Mclver and Wentworth [4]. The RMS
deviation for the 260 EA's was 0.14 e.V. and R2 was 0.98 as shown in Figure 15. The statistical
parameters are shown in Table 2.
30
-------
4 -
'2 —
I -
O
-J -
-3
-2 -I O 1 2: 3 4
O t>scr vccl
Figure 15. SPARC-calculated versus observed electron affinity for 260 organic compounds. The
RMS deviation was 0.14 e.V. and R2 was 0.98.
MONOPOLE MODELS (IONIC SPECIES)
The SPARC models were extended to ionic organic species by incorporating monopole
(charge) electrostatic interaction models to SPARC's physical properties toolbox. These ionic
models play a major role in modeling and estimating Henry's constant for charged (ionic) species in
any solvent system. These capabilities (ionic activity) in turn allow SPARC to calculate gas phase
pKa, and non-aqueous ionization pKa and Ei/2 chemical reduction in any solvent system.
Validation of the SPARC Monopole Models
The SPARC monopole models have been tested on all the available data for Henry's
constant for charged molecules in water, unfortunately there was only 12 data points. However, the
SPARC Ionization pKa in water coupled with Henry's constant for charged molecules was used to
estimate 400 pKa's in the gas phase and 300 pKa's in non-aqueous solvents. Also, SPARC
electron affinity calculator coupled with Henry's constant for charged molecules was used to
-------
estimate 352 £1/2 chemical reduction data measurements. See Table 2 and for more details see
reference 14.
QUALITY ASSURANCE
A quality assurance (QA) plan was developed to recalculate all the aforementioned physical
and chemical properties and compare each calculation to an originally-calculated value stored in
the SPARC databases. Every quarter, two batch files that contain more than 3000 compounds
(4200 calculations) recalculate various physical and chemical properties. QA software compares
every single "new" output to the SPARC originally-calculated-value dating back to 1993-1999.
This ensures the integrity of the SPARC model as new features are added.
CONCLUSION
The strength of the SPARC chemical reactivity parameters and physical properties
calculator is the ability to estimate numerous properties for a wide range of organic compounds
within an acceptable error, especially for molecules that are difficult to measure. The SPARC
physical properties/chemical reactivity parameters calculator prediction is as reliable as most of the
experimental measurements for these properties. For simple structures, SPARC can calculate a
property of interest within a factor of 2 or even better. For complex structures where dipole-dipole
and/or H-bond interactions are strong, properties can generally be calculated within a factor of 3-4.
The true validity of the SPARC physical/chemical property models does not lie in the
models' predictive capability for pKa, or solubility, but is determined by the extrapolatability of
these same models to other types of chemistry. The ability of SPARC models to be extended to
various chemical/physical properties without modification or extra parameterization to any of the
basic models, provides great confidence in this powerful calculation tool.
32
-------
APPENDIX
Summary of usage of the SPARC-web version
Two months back-to-back report, which represents the usage of the SPARC calculator in
October and November, 2002. November was the highest while October was the lowest usage to
date.
Summary of Activity for Report
October 2002
November 2002
Hits Entire Site (Successful) 56,875
Average Number of Hits per day on Weekdays
2,153
Average Number of Hits for the entire
Weekend 1,297
Most Active Day of the Week Thu
Least Active Day of the Week Sat
Most Active Day Ever October 24, 2002
Number of Hits on Most Active Day 4,963
Least Active Day Ever October 05, 2002
Number of Hits on Least Active Day 7
URL's of most active users
207.168.147.52463
pl20xl83.tnrcc.state.tx.us 3,986
141.189.251.7 1,720
198.137.21.14455
57.67.16.50327
gateway.huntingdon.com 6,823
aries.chemie.uni-erlangen.de 1,487
pl20x226.tnrcc.state.tx.us 67
thompson.rtp.epa.gov 413
webcache.crd.GE.COM 143
Hits Entire Site (Successful) 95,447
Average Number of Hits per day on Weekdays
4,146
Average Number of Hits for the entire
Weekend 842
Most Active Day of the Week Wed
Least Active Day of the Week Sun
Most Active Day Ever November 13, 2002
Number of Hits on Most Active Day 15,450
Least Active Day Ever November 02, 2002
Number of Hits on Least Active Day 7
URL's of most active users
141.189.251.7 1,223
gw.bas.roche.com 1,821
gateway.huntingdon.com 3,729
pl20xl83.tnrcc.state.tx.us 737
hwcgate.hc-sc.gc.ca 660
pl20x226.tnrcc.state.tx.us 379
thompson.rtp.epa.gov 563
chen.rice.edu 966
SPARC is online and can be used at http: //ibmlc2 . chem.uqa. edu/sparc
33
-------
REFERENCES
1. S. W. Karickhoff, V. K. McDaniel, C. M. Melton, A. N. Vellino, D. E. Nute, and L. A.
Carreira., US. EPA, Athens, GA, EPA/600/M-89/017.
2. S. W. Karickhoff, V. K. McDaniel, C. M. Melton, A. N. Vellino, D. E. Nute, and L. A.
Carreira., Environ. Toxicol. Chem. 10 1405 1991.
3. S. H. Hilal, L. A. Carreira and S. W. Karickhoff, "Theoretical and Computational
Chemistry, Quantitative Treatment of Solute/Solvent Interactions", Eds. P. Politzer and
J. S. Murray, Elsevier Publishers, chapter 9, 1994.
4. S. H. Hilal, L. A. Carreira, C. M. Melton and S. W. Karickhoff, Quant. Struct. Act. Relat,
12389 1993.
5. S. H. Hilal, L. A. Carreira, C. M. Melton, G. L. Baughman and S. W. Karickhoff, J. Phys.
Org. Chem. 7, 122 1994.
6. S. H. Hilal, L. A. Carreira and S. W. Karickhoff, Quant. Struct. Act. Relat. 14 348 1995.
7. S. H. Hilal, L. A. Carreira, S. W. Karickhoff, M. Rizk, Y. El-Shabrawy and N. A.
Zakhari, Talanta, 43 , 607 1996.
8. S. H. Hilal, L. A. Carreira and S. W. Karickhoff, Talanta., 50 827 1999.
9. S. H. Hilal, L. A. Carreira, S. W. Karickhoff, J. Chromatogr., 269 662 1994.
10. S. H. Hilal, L. A. Carreira and S. W. Karickhoff, Accepted, Quant. Struct. Act. Relat.
11. S. H. Hilal, J.M Brewer, L. Lebioda and L.A. Carreira, Biochem. Biophys. Res. Com., 607
211 1995
12. SAB Report, Evaluation of EPA 's Research on Expert Systems to Predict the Fate and
Effects of Chemicals, November 1991.
13. Peer Review of the Research Programs of the Ecosystems Research Division. U.S. EPA,
NERL, Athens, Ga, June, 1997.
14. S. H. Hilal, a companion U.S. E.P.A report, "Prediction of Chemical Reactivity
Parameters and Physical Properties of Organic Compounds from Molecular Structure
Using SPARC".
34
-------
15. J. E. Lemer and E.Grunwald, Rates of Equilibria of Organic Reactions, John Wiley &
Sons, New York, NY, 1965.
16. Thomas H. Lowry and Kathleen S. Richardson, Mechanism and Theory in Organic
Chemistry. 3ed ed., Harper & Row, New York, NY, 1987.
17. L. P. Hammett, Physical Organic Chemistry, 2nd ed. McGraw Hill, New York, NY, 1970.
18. R.W. Taft, Progress in Organic Chemistry, Vol.16, John Wiley & Sons, New York, NY,
1987.
19. M. J. S. Dewar, The Molecular Orbital Theory of Organic Chemistry, McGraw Hill, New
York, NY, 1969.
20. M. J. S. Dewar and R. C. Doughetry, The PMO Theory of Organic Chemistry, Plenum Press,
New York, NY, 1975.
21. J. Dykyj, M. Repas and J. Anmd Svoboda., Vapor Pressure of Organic substances.
VEDA, Vydavatel Stvo, Slovenskej Akademie Vied, Bratislava, 1984.
22. S. W. Karickhoff and MacArthur Long, US. EPA Internal Report, April 10 1995.
23. C. R. Willke and C. Y Lee, Ind. Eng. Chem. 47 1253 1955.
24. P. D. Neufeld, A. R. Janzen and R. A. Aziz, J. Chem. Phys. 57 1100 1972.
25. S. H. Hilal, L. A. Carreira and S. W. Karickhoff, To be Submitted.
26. R. C. Reid, J. M. Prausnitz and J. K. Sherwood, The Properties of Gases and Liquids, 3ed .,
McGraw-Hill Book Co., 1977.
27. R. R. Dreisbach Physical Properties of Chemical Compounds: Advanced in Chemistry
Series, Dow Chemical Co., ACS, Washington, D.C., (A) Volume I, 1955, (B) Volume H,
1959, (C) Volume III, 1961.
28. R. C. Wilhoit and B. J. Zwolinski, J. Phys. Chem. Ref. Data, 2 1, 1973. Supplement No. 1.
29. T. E. Jordan, The Vapor Pressure of Organic Compounds, Interscience Publisher Inc,
Philadelphia, Pennsylvania, 1954.
30. R. Weast and M. Astle, CRC Handbook of Chemistry and Physics, 79th ed., CRC Press Inc.,
West Palm Beach, 1999.
35
-------
31. D. Mackay, W. Y. Shiu and K.C Ma, Illustrated Handbook of Physical/Chemical Properties
and Environmental Fate of Organic Chemicals, Lewis Publishers, volume I, II, HI, 1993.
32. Douglas Hartley and Hamish Kidd, The Agro Chemical Handbook, Royal Society of
Chemistry, University of Nottingham, England, 1983.
33. S. R. Heller, D.W. Bigwood, P. Laster and K. Scott, The ARSPesticide Properties
Database, Maryland, U.S.A.
34. H. A. Hornsby, D. R. Wauchope and E. Albert Herner, Pesticide Properties in the
Environment, Springer, New York, NY, 1996.
35. E. P. Serjeant and B. Dempsey, lonization Constants of Organic Acids in Aqueous
Solution, Pergamon Press, Oxford, 1979.
36. (A) D. D. Perrin, Dissociation Constants of Organic Bases in Aqueous Solution, Butterwo-
rth & Co, London, 1965 & Supplement 1972. (B) Supplement 1972.
37. N. B. Chapman, J. Chem. Soc., 1291 1963.
38. L. W. Deady and R. A. Shanks, Aust. J. Chem., 25, 2363 1972.
39. M. L. Bender and Robert J. Thomas, J. Am. Chem. Soc., 83 4189 1961.
40. DeLos DeTar and Carl J. Tenpas, J. Am. Chem. Soc., 7903 1976.
36
------- |