Physical Property Prediction in Organic Chemistry

An important factor in technological or scientific innovation is the ready availability of relevant information. In chemical research and development data and factual information on chemical species is of prime importance: in chemistry at present we know of about 9 Mil. compounds — this figure is increased annually as a result of the synthesis of another half a million substances. New compounds, new properties, new processes and new methods were reported in the last year in ca. 600 000 articles, with an upward trend predicted for the future. The state-of-the-art is documented in around 40 000 patent applications annually, more than 16% of which are from the Federal Republic of Germany.

The Beilstein Information System is a structure oriented collection of physical data and reactions which have been extracted from primary literature. The Beilstein Handbook contains more than 1 million compounds described in over 340 volumes. This information is available in printed form for more than 100 years. The Beilstein-Institute is currently setting up a numerical factual databank based on these 340 volumes of the Beilstein Handbook of Organic Chemistry (literature from 1830 to 1960) and 7.5 million factual records of organic compounds (literature from 1960 to 1980).

“Has Beilstein got enough data ?” The answer to this question depends on what the data are used for.

Questions and issues about property prediction are addressed and discussed. Issues such as reliability, evaluation of predicted values and how to handle different/multiple values predicted from different methodology, and how the data should be presented to the user community are critical to a project of the scale of the Beilstein database of chemicals. Lastly a proposal is presented as to how to initially address large-scale prediction of properties of chemicals.

Beilstein’s handbook of organic chemistry is currently implemented as a database on STN, starting with the first part containing the heterocyclic substances.It is the intention of this paper to give an overview of the capabilities of this database on STN with special focus on the numeric features. At first, an introduction to the database is presented starting with an outline of the database design. Following, the various possibilities to search and retrieve the documents from the Beilstein database are discussed. In addition to the normal search for the content of a field it is also possible to search for the name of the field, thus offering an alternative way to access the file.Several new software features are required to support especially the numeric fields of this database. Most numeric properties are associated with an experimental uncertainty, which must be taken into account by the database loading and retrieval software. This new feature is called numeric range search capability. The problem is briefly described together with a possible solution. In general, a numeric interval can be entered as a range with a lower limit and an upper limit. However, many scientists are more familiar with the notation of a measured value and its associated uncertainty. A feature which allows the specification of a range in terms of a value plus/minus a tolerance offers additional user support. Another important function for numeric databases is the conversion of physical units. Especially in the older literature many units are found which are no longer used mainly due to a standardization process. To overcome the difficulties which are inherent to the use of different units, a function is required which allows the user to work with his own set of physical units and lets the software worry about the transformation to the units used in the specific field of this database. At last, some future developments and requirements are outlined.

The Thermodynamics Research Center (TRC) maintains several large collections of thermodynamic and physical property data on organic compounds. This data is being transferred to computer accessible databases. The TRC Source File contains measured data from the scientific literature including citations to sources of data, compounds names, CA registry numbers, sample descriptions, numerical data, uncertainty estimates and quality codes for most thermophysical properties. The TRC Selected Data File contains values of selected properties (either evaluated or estimated) published in the TRC Thermodynamic Tables. This database will be used for the automatic generation of camera-ready hard copies. Both of these databases will be described.

The United States National Cancer Institute manages a program whose goal is the discovery and development of drugs to be used in the treatment of cancer. This Program, which began in 1955, has examined approximately 500,000 materials for anticancer activity and it continues to screen much smaller numbers of compounds per year. The large database that has been built as a result of this work was used between 1980 and 1985 as the basis of an experiment in which the potential biological activity of a chemical structure was estimated before the compound was acquired and tested. The accuracy of such estimates and the impact upon the overall program is described.

A knowledge of physico-chemical properties is a fundamental requirement of all design activities in the chemical industry. Experimental values comprise only a small part of the data required for further application. Estimation and prediction methods are in many cases the only way to get the information needed.

This article aims to provide a short review of the methods available for the calculation of molecular structures and energies and other properties. The programs employed all use a guessed initial geometry as starting point for the calculation and then optimize the structure in order to find the minimum that can be found by moving down in energy from the starting point. This leads to the first problem with such methods, especially for very large molecules. There are at present very few programs that can investigate a series of possible structures in order to identify as many minima as possible and to be able to find the global (most stable) minimum within any degree of certainty. Even when this is possible (at present only for molecular mechanics calculations) the cost in computer time can be very large. For the other methods, the chemist must have enough imagination to be able to predict all the possibilities open to the molecule in order to find the global minimum.

About one hundred years ago, the relationship between the boiling point of a substance and its heat of vaporization at the boiling point (Trouton’s Rule) was reported in the chemical literature. Since that time, much progress has been made in the correlation of physical and thermodynamic properties. Physical organic chemists in particular have been active in the development of correlation and estimation schemes linking the molecular structures of organic compounds and various thermodynamic properties. Some correlation and estimation schemes are more successful than others. Of the estimation schemes developed, the one put together by S.W. Benson and coworkers has had more universal acceptance because of its overall simplicity, ease of application, and general good agreement between estimated and experimental values. This scheme assigns individual group energy values for molecular fragments which are additive, account for nearest neighbor interactions, and give special consideration to corrections for steric strain or stereoisomerism. The primary focus of this scheme has been on organic molecules in the gas phase, although some applications to the liquid and solid phases have been reported.Our recent development of the Benson approach to the estimation of thermodynamic properties (enthalpy of formation, heat capacity, and entropy) at 298.15 K has focused upon the liquid and solid phases. The gas phase has been included also for the sake of continuity and internal consistency of the calculation of thermodynamic properties among the three phases. Studies showing the predictive capability of this scheme toward estimating the thermodynamic properties of hydrocarbons and organic compounds containing the elements: CHO, CHN, and CHNO have already been carried out. Work is in progress to complete the scope of the predictive capability to cover organic compounds containing the elements: sulfur, phosphorus, fluorine, chlorine, bromine, iodine, and metals.This paper provides a general overview of the NBS program in the estimation of the thermodynamic properties of organic compounds using the Benson approach and presents discussions of selected topics, such as: (1) relationships of the enthalpy of formation, heat capacity, and entropy at 298.15 K to other thermodynamic properties (Gibbs energy of formation, equilibrium constants, enthalpies and entropies of transition) which extend predictive capabilities, (2) descriptions of the Benson notation, group values and their application to the estimation of thermodynamic properties, and (3) explanations of unique solutions to the estimation of two selected classes of organic compounds.

Methods have been developed that allow the calculation of energetic and electronic effects in organic molecules. The values thus obtained can be used for the prediction of a variety of physical and chemical data of organic compounds.

Statistical thermodynamics provides a simple formal connection between the thermodynamics of a system, as represented by its free energy A, and the molecular properties of a system, as represented by its canonical partition function Q [1]: (1)$${{\rm{A}}^{{\rm{(T,V,\{ }}{{\rm{N}}_{\rm{j}}}{\rm{\} )}}}}\,{\rm{ = }}\,{\rm{ - }}\,{\rm{kT}}\,{\rm{ln}}\,{\rm{Q(T,V,\{ }}{{\rm{N}}_{\rm{j}}}{\rm{\} ),}}$$ where k is Boltzmann’s constant, T is the thermodynamic temperature, V the volume, {N_j} the total amount of molecule numbers of the various components, and (2)$${\rm{Q}}\,{\rm{ = }}\,\mathop {\rm{\Sigma }}\limits_{\rm{i}} \,{\rm{e}}{\,^{{\rm{ - E}}{}_{\rm{i}}{\rm{/kT,}}}}$$ Here E_i, the molecular energy of the system in its quantum state i, is the key quantity.

Given the interpolation points (x_i,f_i) (“knots”) we search for coefficients a_i satisfying the interpolation condition [6]: (1)$${\rm{I(}}{{\rm{a}}_{\rm{o}}}{\rm{, \ldots ,}}{{\rm{a}}_{\rm{n}}}\,{\rm{;}}\,{{\rm{x}}_{\rm{i}}}{\rm{)}}\,{\rm{ = }}\,{{\rm{f}}_{{\rm{i}}\,{\rm{,}}\,}}\,{\rm{i}}\,\,{\rm{ = }}\,\,{\rm{0, \ldots ,m}}\,\,{\rm{(m}}\,{\rm{ = }}\,{\rm{n),}}\,{\rm{where}}\,{{\rm{x}}_{\rm{i}}}\, \in \,{{\rm{R}}^{\rm{N}}}\,{\rm{and}}\,{{\rm{f}}_{\rm{i}}}\, \in \,{\rm{R}}\,{\rm{(R}}\,{\rm{ = }}\,{\rm{real}}\,{\rm{space)}}{\rm{.}}$$

Estimation of missing data in an incomplete data base is possible by means of physico-chemical (deterministic) models or by using inherent information in the data base under study. The use of deterministic modelling techniques is limited to those objects (compounds) that fit within all the necessary assumptions and therefore this method will be applicable to a subset of compounds only or several models have to be created in combination with the data base.

This paper summarises an extended research programme to investigate the use of fragment-based measures of inter-molecular similarity in chemical information systems, with particular reference to structure-property correlation. Comparative studies are reported of structural similarity measures and of clustering methods for chemical structure databases. The methods are most appropriate when very sparse data matrices are available; in such cases, a very fast nearest neighbour searching algorithm can be used for the calculation of the requisite similarities.

Relationships between molecular structure and biological activity or molecular structure and physical properties can be investigated for large sets of organic compounds using computer-assisted methods. Our research involves the design, implementation, testing, and application of computer software for the purpose of discovering structure-property and structure-activity relationships and thus developing the capability to predict properties for unknown compounds. The approach involves the graphical entry and storage of structures, three-dimensional molecular modeling, molecular structure descriptor generation, and analysis of the descriptors using pattern recognition methods or multivariate statistical methods. The computer-generated structural descriptors represent the molecules topologically (e.g., path counts, molecular connectivity), geometrically (e.g., molecular volume, surface area, principal moments), electronically (e.g., partial charges, bond orders), and physicochemically (e.g., log P, molar refractivity). A large, fully-integrated, interactive software system, called ADAPT for Automated Data Analysis and Pattern recognition Toolkit, has been developed to make such S AR and SPR research convenient. ADAPT is under continual development through the introduction of new molecular structure descriptors and new analysis methods. A number of successful studies have been reported in property prediction (prediction of boiling points of olefins, GC and HPLC retention indices, and simulation of 13C NMR chemical shifts) and in the structure-activity area (pharmaceutical drugs, olfactory stimulants, mutagens, carcinogens, anti-tumor drugs). Examples of current studies include: S AR of anti-tumor retinoids, carcinogenicity of N-nitroso compounds, HPLC retention indices of PACs and the importance of molecular shape, GC retention indices of polychlorinated biphenyls, 13C NMR simulation of substituted norbornanes.

Quantitative structure activity relationships (QSAR) are to be understood as a consequence of the fact that the interactions of drugs with their biological counterparts are determined by intermolecular forces. Thus the structural dependence of biological activities can be described either by physicochemical parameters (Hansch analysis) or by indicator variables encoding different structural features (Free Wilson analysis, pattern recognition).

Molecules, due to their electron density distribution, are conveniently described as entities with a finite extension in physical space, i.e. as solid bodies. This is reflected in the popular CPK plastic models that almost every chemist has used for educational or visualization purposes.

Research and development of molecular design support systems has been carried out actively in Japan as well as in the U. A. and Europe. All systems proposed so far, however, are designed only for management of data on chemical compounds or for statistic or various molecular calculations. The authors have been engaged in development of a comprehensive molecular design support system, called TUTORS1), that, in addition to serving for the above- mentioned purposes, can generate and propose candidate structures of new promising compounds that are expected to have specific functions. This is the ultimate goal of studies in the field of molecular design. Some unique techniques are also developed and incorporated. The present report describes the present status of the development of the TUTORS system, centering on the basic concept of computer-assisted molecular design and major functions of the system.

The environmental fate and mobility assessment is an essential part of the process of selecting and identifying environmental chemicals. However, for most chemicals only fragmentary knowledge exists about those properties which determine their fate in the environment. Therefore, it is pertinent to estimate these physical-chemical data on chemicals. Due to the huge number of organics not only an appropriate package of formula but also a high degree of automatism for this task is needed.There are two types of approaches to estimate properties. The first one is to use a databasis for fragments of the molecule. The second one is to use property-property- relationships which in general do not need the knowledge of (sub)structures.Therefore, DTEST which is part of the EDP-code E4CHEM (acronym for “Exposure and Ecotoxicity Estimation for Environmental Chemicals”) mainly property-property- relationships are used which allow a high degree of automatism. The relationship to calculate vapor pressure, partition coefficients, solubility and other environmental relevant properties are shown and discussed in DTEST. A brief explanation of the leading ideas in the program structure is given.

The chemical industry has grown enormously in recent decades. It provides us with numerous vital chemicals (fuels, antibiotics and other drugs, plastics, pesticides, fertilizers, etc.) without which our society cannot survive and preserve its present life style and high living standards. Many of these substances have little or no adverse environmental effects, but some may be harmful to human health and the natural environment. Usually these effects only become apparent after wide and prolonged usage and at that point authorities introduce control measures. Clearly, there is a need for an effective evaluation and testing program to identify, before their use, those chemicals or classes of chemicals that present a potential environmental hazard. Such evaluation procedure should trace the fate of chemicals from discharge and dispersal to subsequent effects on biota. The ecotoxicological profile of a chemical is based on a sequence of interactions and effects controlled by its physical, chemical, and biological properties. At the first stage, a chemical released into the environment is subject to physical distribution between the atmosphere (air), water, soils, and sediment depending on its physico-chemical properties. At the same time, it can be chemically modified and degraded by abiotic processes or more often by microorganisms in the environment. During the following stage organisms will be exposed to the chemical either in its original or in its degraded or transformed form. The uptake of the chemical and degradation products will occur. Organisms may react to such exposure by variety of negligible and sublethal effects or ultimately by death.

Estimation methods for physical properties are widely used in industry. The number of chemical substances is enormous (several million) and thus complete or even partial experimental data are rare. In addition, experimental measurements are expensive and thus screening studies and preliminary design often must be done without measured properties.

The PFGC group-contribution equation of state as an example of a promising method to solve industrial application problems for complex mixtures containing non-polar, polar, associating, sub- and supercritical components, has been used to demonstrate the following aspects of general interest: The industrial requirements for improved physical properties prediction methods, and the range of molecular com, lexity of mixtures encountered in gas purification processes, are discussed.A general strategy for the preselection of physical property prediction methods, and especially those for testing equations of state or group- contribution equations of state, has been developed. The industrial boundary conditions which determine success or failure are given.A general multiproperty-multicomponent-multiphase strategy to be used for the fit of the parameters of equations of state or group-contribution equations is proposed.For the PFGC group-contribution equation of state the investigations have the following specific results: The PFGC method has many fundamental weaknesses.The original PFGC method does not work for industrial applications with sufficient accuracy.From an industrial point of view, a group-contribution equation of state is still required for all types of molecular interactions and all areas of applications.

In the last 20 years great progress has been achieved in the calculation and prediction of the real behavior of multicomponent systems using gE-models or EOS and binary information alone.When the required binary data are missing, the group contribution method UNIFAC can be applied to complete the available experimental information; however, this method also shows some weaknesses.In this paper results for the UNIFAC method will be presented together with the improvements which are obtained when a modified form of the UNIFAC method and a larger data base are used.

Thermochemical data has been collected, processed and assessed at the University of Sussex over a period of years. Particular emphasis has been placed on standard enthalpies of formation of organic compounds, culminating in 1986 in a book containing values for approximately 3000 compounds of the elements C, H, O, N, S and halogens. Since then, data on entropies and heat capacities have been acquired and all the data set up as computer files which may be searched for compounds containing or not containing specific structural features. Software has also been developed for calculating values of thermochemical properties using files of parameters defined to fit those experimental values which are considered to be most reliable.

Considerable effort has been directed in the past toward understanding the relationship between molecular structure and the distribution-transport characteristics of chemicals and drugs. Although distribution is an equilibrium process and transport is a rate process, transport is subject to distribution for all passive diffusion processes and for most active transport phenomena in their presaturation phase. For example, a substantial portion of drug absorption involves partitioning into the rate-controlling membrane. The lipophilic-hydrophilic balance of ophthalmic drugs, for instance, largely determines their transcorneal absorption rate. Similarly, the rate of GI absorption of passively absorbed drugs usually depends on their ability to partition into the rate-determining barrier membrane. This condition is also true for dermal agents.

In spite of its importance to a number of scientific disciplines, there are few reliable techniques for the estimation of the aqueous solubility of a crystalline organic compound. One of the major reasons for the lack of progress in this field is the fact that solubility is not strictly an additive, constituative property, and thus cannot be predicted by group contributions alone.

Many problems of practical interest require the knowledge of the equilibrium properties of multicomponent mixtures. With the availability of modern computers complex semiempirical models for the description of the real behavior of these mixtures became popular. These models typically require two or three parameters per binary pair of components, which have to be fitted to experimental mixture data.Although the description of the real behavior as a function of the liquid phase composition is mostly of good accuracy, the parameters used are only valid for a limited temperature region.This paper describes the structure of an interactive program package for the simultaneous correlation of different excess properties.Results for the system ethanol-water are presented, which stress the importance of temperature-dependent interaction parameters.The program-package RECVAL is directly interfaced to the Dortmund Data Bank (DDB), makes excessive use of graphics for data representation and may be extended for the correlation and prediction of pure component properties.

The aqueous solubility of organic compounds is important in a large number of areas of science. For this reason, solubility is a frequently measured property. Solubilities are reported in scientific publications representing nearly every scientific discipline. Yet in spite of the tremendous amount of solubility data that is available in the literature, it is difficult to find the solubility of a particular compound from a normal literature search. When values are available in the literature there are often major discrepancies among the values reported for a particular compound by different authors and it is difficult to select the best value from among all of the reported values. Hence, the ARIZONA dATAbASE was started.

Vapour pressures are among the most often measured physico-chemical properties. The measurements are most frequent and accurate in the pressure range 10–100 kPa. The determination of vapour pressures in the region below 1 kPa, on the other hand, are difficult, time-consuming and relatively inaccurate. Extrapolation based on vapour pressures measured in a limited range near the boiling point is usually unreliable.

Vapor pressure data are abundant and accurate near the normal boiling point while they are scarce and unreliable in the low pressure range. Data on thermal properties (enthalpy of vaporization and difference in the heat capacities of an ideal gas and the liquid) are available, however at temperatures far below the normal boiling point. All three properties are related by exact thermodynamic relationships and can be correlated simultaneously. This procedure can be used in the evaluation of thermodynamic data on vaporization equilibria as a rigorous consistency test and for producing recommended data sets. When a suitable correlation equation is selected a single set of parameters permits generation of consistent data for several properties between the triple and normal boiling points. The procedure is especially useful for calculating vapor pressures and/or enthalpies of vaporization far below the normal boiling point. In combination with the group contribution methods for estimation of thermal properties, the principle of simultaneous correlation can serve as a base in formulation of a new approach to the prediction of vapor pressures and/or enthalpies of vaporization at low reduced temperatures. Such a procedure would not require use of any traditional parameters (critical constants, acentric factor, etc.) which are not available for most high boiling compounds.

Several compilations of heat capacities of liquids have been published over the last 20 years. However, some compilations are limited with regards to the number of compounds included, some by the degree of critical evaluation of the data, and some in the temperature range which the data cover. Besides these compilations, there are others which contain a number of tables having heat capacities for industrially important substances and other related properties. These multi-property tables mostly contain old data, with unclear identification of the sources and sometimes questionable levels of accuracy. Most of these tables lack a critical approach in the selection and evaluation of the data.

The progress in the field of chemistry as a scientific discipline, especially in this century, has consisted in setting up models, concepts and rules so as to create order and understanding for the enormous amount of chemical observations and factknowledge regarding connectivity and reactions.

The view of the overwhelming majority of participants, including the Beilsteiners, was that the workshop was a resounding success. This success was not just a measure of the excellent food and accommodation provided by the Hotel Schloss Korb, but was also a reflection of the content of the scientific programme. The wish expressed by all for another workshop could not be correlated with the weather, since due to the incessant rain we were virtual (albeit willing) prisoners in a real castle.