Machine learning transition temperatures from 2D structure.

A priori knowledge of physicochemical properties such as melting and boiling could expedite materials discovery. However, theoretical modeling from first principles poses a challenge for efficient virtual screening of potential candidates. As an alternative, the tools of data science are becoming increasingly important for exploring chemical datasets and predicting material properties. Herein, we extend a molecular representation, or set of descriptors, first developed for quantitative structure-property relationship modeling by Yalkowsky and coworkers known as the Unified Physicochemical Property Estimation Relationships (UPPER). This molecular representation has group-constitutive and geometrical descriptors that map to enthalpy and entropy; two thermodynamic quantities that drive thermal phase transitions. We extend the UPPER representation to include additional information about sp2-bonded fragments. Additionally, instead of using the UPPER descriptors in a series of thermodynamically-inspired calculations, as per Yalkowsky, we use the descriptors to construct a vector representation for use with machine learning techniques. The concise and easy-to-compute representation, combined with a gradient-boosting decision tree model, provides an appealing framework for predicting experimental transition temperatures in a diverse chemical space. An application to energetic materials shows that the method is predictive, despite a relatively modest energetics reference dataset. We also report competitive results on diverse public datasets of melting points (i.e., OCHEM, Enamine, Bradley, and Bergström) comprised of over 47k structures. Open source software is available at https://github.com/USArmyResearchLab/ARL-UPPER.

Estimation of Melting Points of Organics.

Unified physicochemical property estimation relationships is a system of empirical and theoretical relationships that relate 20 physicochemical properties of organic molecules to each other and to chemical structure. Melting point is a key parameter in the unified physicochemical property estimation relationships scheme because it is a determinant of several other properties including vapor pressure, and solubility. This review describes the first-principals calculation of the melting points of organic compounds from structure. The calculation is based on the fact that the melting point, Tm, is equal to the ratio of the heat of melting, ΔHm, to the entropy of melting, ΔSm. The heat of melting is shown to be an additive constitutive property. However, the entropy of melting is not entirely group additive. It is primarily dependent on molecular geometry, including parameters which reflect the degree of restriction of molecular motion in the crystal to that of the liquid. Symmetry, eccentricity, chirality, flexibility, and hydrogen bonding, each affect molecular freedom in different ways and thus make different contributions to the total entropy of fusion. The relationships of these entropy determining parameters to chemical structure are used to develop a reasonably accurate means of predicting the melting points over 2000 compounds.

A rule of unity for human intestinal absorption 3: Application to pharmaceuticals.

The rule of unity is based on a simple absorption parameter, Π, that can accurately predict whether or not an orally administered drug will be well absorbed or poorly absorbed. The intrinsic aqueous solubility and octanol-water partition coefficient, along with the drug dose are used to calculate Π. We show that a single delineator value for Π exist that can distinguish whether a drug is likely to be well absorbed (FA ≥ 0.5) or poorly absorbed (FA < 0.5) at any specified dose. The model is shown to give 82.5% correct predictions for over 938 pharmaceuticals. The maximum well-absorbed dose (i.e. the maximum dose that will be more than 50% absorbed) calculated using this model can be utilized as a guideline for drug design and synthesis.

Estimating the Physicochemical Properties of Polysubstituted Aromatic Compounds Using UPPER.

The UPPER model (Unified Physicochemical Property Estimation Relationships) has been used to predict 9 essential physicochemical properties of pure compounds. It was developed almost 25 years ago and has been validated by the Yalkowsky group for almost 2000 aliphatic, aromatic, and polyhalogenated hydrocarbons. UPPER is based on a group of additive and nonadditive descriptors along with a series of well-accepted thermodynamic relationships. In this model, the 2-dimensional chemical structure is the only input needed. This work extends the applicability of UPPER to hydrogen bonding and non-hydrogen bonding aromatic compounds with several functional groups such as alcohol, aldehyde, ketone, carboxylic acid, carbonate, carbamate, amine, amide, nitrile as well as aceto, and nitro compounds. The total data set includes almost 3000 compounds. Aside from the enthalpies and entropies of melting and boiling, no training set is used for the calculation of the properties. The results show that UPPER enables a reasonable estimation of all the considered properties.

Imaging of in vitro parenteral drug precipitation.

Solid particulate matter introduced into the bloodstream as a result of parenteral drug administration can produce serious pathological conditions. Particulate matter that cannot be eliminated by pre-infusion filtration is often the result of drug precipitation that occurs when certain parenteral formulations are mixed with blood. A new device is designed to model the mixing of drug formulations with flowing blood utilizing a uniquely designed flow cell and a CCD camera to view the formulation as it is mixed with a blood surrogate in real time. The performance of the proposed device is measured using 3 commercially available parenteral formulations previously tested using a validated in vitro model.

Estimating the physicochemical properties of polyhalogenated aromatic and aliphatic compounds using UPPER: part 2. Aqueous solubility, octanol solubility and octanol-water partition coefficient.

The UPPER (Unified Physicochemical Property Estimation Relationships) model uses additive and non-additive parameters to estimate 20 biologically relevant properties of organic compounds. The model has been validated by Lian and Yalkowsky (2014) on a data set of 700 hydrocarbons. Recently, Admire et al. (2014) expanded the model to predict the boiling and melting points of 1288 polyhalogenated benzenes, biphenyls, dibenzo-p-dioxins, diphenyl ethers, anisoles and alkanes. In this work, 19 new group descriptors are determined and used to predict the aqueous solubilities, octanol solubilities and the octanol-water coefficients.

The rule of unity for human intestinal absorption 2: application to pharmaceutical drugs that are marketed as salts.

The efficiency of the human intestinal absorption (HIA) of the 59 drugs which are marketed as salts is predicted using the rule of unity. Intrinsic aqueous solubilities and partition coefficients along with the drug dose are used to calculate modified absorption potential (MAP) values. These values are shown to be related to the fraction of the dose that is absorbed upon oral administration in humans (FA). It is shown that the MAP value can distinguish between drugs that are poorly absorbed (FA <0.5) and those that are well absorbed (FA ≥ 0.5). Inspection of the data as well as a receiver operative characteristic (ROC) plot shows that a single critical MAP value can be used for predicting efficient human absorption of drugs. This forms the basis of a simple rule of unity based solely on in vitro data for predicting whether or not a drug will be well absorbed at a given dose.

Estimating the physicochemical properties of polyhalogenated aromatic and aliphatic compounds using UPPER: part 1. Boiling point and melting point.

The UPPER (Unified Physicochemical Property Estimation Relationships) model uses enthalpic and entropic parameters to estimate 20 biologically relevant properties of organic compounds. The model has been validated by Lian and Yalkowsky on a data set of 700 hydrocarbons. The aim of this work is to expand the UPPER model to estimate the boiling and melting points of polyhalogenated compounds. In this work, 19 new group descriptors are defined and used to predict the transition temperatures of an additional 1288 compounds. The boiling points of 808 and the melting points of 742 polyhalogenated compounds are predicted with average absolute errors of 13.56 K and 25.85 K, respectively.

Carnelley's rule and the prediction of melting point.

Carnelley (1882) made some important and useful observations on the relationship between the arrangement of the atoms in a molecule and its melting point. According to Brown and Brown (2000. J Chem Ed 77:724-731), Carnelley's rule states "of two or more isometric compounds, those whose atoms are the more symmetrically and the more compactly arranged melt higher than those in which the atomic arrangement is asymmetrical or in the form of long chains." Carnelley's rule can best be understood and quantitated from the dependence of the entropy of melting upon molecular geometry.

Unified physicochemical property estimation relationships (UPPER).

The knowledge of physicochemical properties of organic compounds becomes increasingly important in pharmaceutical sciences, chemical engineering, and other fields. In this study, we developed UPPER (Unified Physicochemical Property Estimation Relationships), a comprehensive model for the estimation of 20 physicochemical properties of organic compounds. UPPER is a system of thermodynamically sound relationships that relate the various phase-transition properties to one another, which includes transition heats, transition entropies, transition temperatures, molar volume, vapor pressure, solubilities and partition coefficients in different solvents, and so on. UPPER integrates group contributions with the molecular geometric factors that affect transition entropies. All of the predictions are directly based on molecular structure. As a result, the proposed model provides a simple and accurate prediction of the properties studied. UPPER is designed to predict industrially, pharmaceutically, and environmentally relevant physicochemical properties. It can be an aid for the efficient design and synthesis of compounds with optimal physicochemical properties.

Predicting the octanol solubility of organic compounds.

The molar octanol solubility of an organic nonelectrolytes can be reasonably predicted solely from its melting point provided that its liquid (or a hypothetical super-cooled liquid) form is miscible with octanol. The aim of this work is to develop criteria to determine if the real or hypothetical liquid form of a given compound will be miscible with octanol based on its molar volume and solubility parameter. Fortunately, most organic compounds (including most drugs) conform to the criteria for complete liquid miscibility, and therefore have solubilities that are proportional to their melting points. The results show that more than 95% of the octanol solubilities studied are predicted with an error of less than 1 logarithmic unit.

Preformulation study of NSC-726796.

A stability-indicating high-performance liquid chromatography method to quantify 2-(2,4-difluorophenyl)-4,5,6,7-tetrafluoroisoindoline-1,3-dione (NSC-726796) and its three main degradation products was developed. This method was used to investigate its degradation kinetics and mechanism. The reaction follows first-order kinetics and appears to be base catalyzed with the maximum stability at pH 1. The products were identified as 2-(2,4-difluorophenylcarbamoyl)-3,4,5,6-tetrafluorobenzoic acid (NSC-749820), 2,4-difluoroaniline, and tetrafluorophthalic acid. The parent drug, NSC-726796, was also found to react with methanol and ethanol. NSC-726796 demonstrates antiangiogenic activity, however, when its degradant NSC749820 does not show antiangiogenic activity.

Perspective on improving passive human intestinal absorption.

Methods such as pH adjustment, cosolvency, complexation, and micellization are routinely used to increase the concentration of dissolved drug in the gastrointestinal (GI) lumen over that of a saturated solution. However, these solubilizing agents also reduce the membrane-water distribution coefficient so that the membrane transport rate is not changed. Also, dilution of a formulation upon administration results in: (1) a pH change toward that of the GI fluid, (2) an exponential decrease in cosolvency, and (3) disassociation of complexes and the disintegration of micelles. As a result, these solubilizing agents cannot be expected to produce any increase in membrane transport-limited drug absorption over that of a suspension of unformulated drug.

Stability of 5-fluoro-2'-deoxycytidine and tetrahydrouridine in combination.

In vivo, the DNA methyltransferase inhibitor, 5-fluoro-2'-deoxycytidine (FdCyd, NSC-48006), is rapidly converted to its unwanted metabolites. Tetrahydrouridine (THU, NSC-112907), a cytidine deaminase inhibitor can block the first metabolic step in FdCyd catabolism. Clinical studies have shown that co-administration with THU can inhibit the metabolism of FdCyd. The National Cancer Institute is particularly interested in a 1:5 FdCyd/THU formulation. The purpose of this study was to investigate the in vitro pH stability of FdCyd and THU individually and in combination. A stability-indicating high-performance liquid chromatography method for the quantification of both compounds and their degradants was developed using a ZIC(R)-HILIC column. The effect of THU and FdCyd on the in vitro degradation of each other was studied as a function of pH from 1.0 to 7.4 in aqueous solutions at 37 degrees C. The degradation of FdCyd appears to be first-order and acid-catalyzed. THU equilibrates with at least one of its degradants. The combination of FdCyd and THU in solution does not affect the stability of either compound. The stability and compatibility of FdCyd and THU in the solid state at increased relative humidity and at various temperatures are also evaluated.

Estimation of the ideal solubility (crystal-liquid fugacity ratio) of organic compounds.

Melting point, entropy of melting and heat capacity of melting are required for the calculation of the ideal solubility of a solid solute via the Clausius-Clapyron equation. This article reviews the published approximations of estimating entropy and heat capacity of melting. By comparing the available experimental results to calculated values the authors attempt to identify the best estimation of the ideal solubility and crystal-liquid fugacity ratio for organic compounds.

Prediction of aqueous solubility from SCRATCH.

This study proposes the SCRATCH model for the aqueous solubility estimation of a compound directly from its structure. The algorithm utilizes predicted melting points and predicted aqueous activity coefficients. It uses two additive, constitutive molecular descriptors (enthalpy of melting and aqueous activity coefficient) and two non-additive molecular descriptors (symmetry and flexibility). The latter are used to determine the entropy of melting. The melting point prediction is trained on over 2200 compounds whereas the aqueous activity coefficient is trained on about 1640 compounds, making the model very rigorous and robust. The model is validated using a 10-fold cross-validation on a dataset of 883 compounds for the aqueous solubility prediction. A comparison with the general solubility equation (GSE) suggests that the SCRATCH predicted aqueous solubilities have a slightly greater average absolute error. This could result from the fact that SCRATCH uses two predicted parameters whereas the GSE utilizes one measured property, the melting point. Although the GSE is simpler to use, the drawback of requiring an experimental melting point is overcome in SCRATCH which can predict the aqueous solubility of a compound based solely on its structure and no experimental values.

An interesting relationship between drug absorption and melting point.

The ability to predict the extent of passive intestinal drug absorption is very important for efficient lead candidate selection and development. Physicochemical-based absorption predictive models previously developed use solubility, partition coefficient and pK(a) as drug input parameters for intestinal absorption. Alternatively, this study looks at the relationship between melting point and passive transport for poorly soluble drugs. It is based entirely on the expression derived from the General Solubility Equation (GSE) that relates melting point to the product of intrinsic solubility and partition coefficient. Given that the melting point of a compound is one of the first and more reliable physical properties measured, it can be advantageously used as a guide in early drug discovery and development. This paper elucidates the interesting relationship between the melting point and dose to the fraction absorbed of poorly soluble drugs, i.e., class II and IV compounds in the Biopharmaceutics Classification System. The newly defined melting point based absorption potential (MPbAP) parameter is successful at distinguishing 90% of the 91 drugs considered being well absorbed (FA>0.5) or poorly absorbed. In general, lower melting compounds are more likely to be well absorbed than higher melting compounds for any given dose. The fraction absorbed for drugs with high melting temperatures is limited by the dose to a greater degree than it is for low melting compounds.

Preformulation and pharmacokinetic studies on antalarmin: a novel stress inhibitor.

The preformulation, solubilization and pharmacokinetic evaluation of antalarmin, a stress inhibitor, have been conducted. Antalarmin has a poor water solubility of less than 1 microg/mL and is weakly basic with an experimentally determined pK(a) of 5.0. Multiple solubilization approaches including pH-control either alone or in combination with cosolvents, surfactants and complexing agents have been investigated. The applicability of lipid-based systems has also been explored. Four formulations, each with a targeted drug loading capacity of 100 mg/mL, show potential for oral administration. Three of these formulations are aqueous solutions (10% ethanol + 40% propylene glycol; 20% cremophor EL; 20% sulfobutylether-beta-cyclodextrin) each buffered at pH 1. The fourth formulation is a lipid-based formulation comprising of 20% oleic acid, 40% cremophor EL and 40% Labrasol. No precipitation was observed following dilution of the four formulations with water and enzyme free simulated gastric fluid. However, only the lipid-based formulation successfully resisted drug precipitation following dilution with enzyme free simulated intestinal fluid. Pharmacokinetic analysis conducted in rats revealed that the 20% cremophor EL solution formulation has a fivefold higher oral bioavailability compared to a suspension formulation. The lipid-based formulation resulted in over 12-fold higher bioavailability as compared to the suspension formulation, the highest amongst the formulations examined.

Predicting aqueous solubility: the role of crystallinity.

This paper revisits the role of crystallinity in predicting the aqueous solubility of a wide variety of organic compounds. Box and Comer (Current Drug Metabolism, 2008, 9, 869-878) fitted solubility data for 86 drugs to an equation based solely on log P. The General Solubility Equation of Jain and Yalkowsky, which accounts for the crystal lattice energy, was applied to the same data set and gives more accurate solubility predictions. In this simple comparison between two solubility prediction methods, we show that log P alone is only half of the solution, and that there is a need to include the melting point when dealing with crystalline solutes.

2-[N-(2,4-Difluoro-phen-yl)carbamo-yl]-3,4,5,6-tetra-fluoro-benzoic acid.

The title compound, C(14)H(5)F(6)NO(3), was synthesized by condensation of tetra-fluoro-phthalic anhydride and 2,4-difluoro-aniline. It was then recrystallized from hexane to give a nonmerohedral twin with two crystallographically unique mol-ecules in the asymmetric unit. The refined twin fraction is 0.460 (3). Torsional differences between the aryl rings and the central amide group account for the presence of two unique mol-ecules. The compound packs as double tapes formed by O-H⋯O and N-H⋯O hydrogen-bonding inter-actions between each unique mol-ecule and its symmetry equivalents.