Aqueous Solvation Free Energies Research Articles

Density Functional Theory with Implicit Solvents for Accurate Estimation of Aqueous and Organic Solvation Free Energies of Lignin Fragments

Density Functional Theory with Implicit Solvents for Accurate Estimation of Aqueous and Organic Solvation Free Energies of Lignin Fragments

Liquid-Phase Effects on Adsorption Processes in Heterogeneous Catalysis.

Aqueous solvation free energies of adsorption have recently been measured for phenol adsorption on Pt(111). Endergonic solvent effects of ∼1 eV suggest solvents dramatically influence a metal catalyst’s activity with significant implications for the catalyst design. However, measurements are indirect and involve adsorption isotherm models, which potentially reduces the reliability of the extracted energy values. Computational, implicit solvation models predict exergonic solvation effects for phenol adsorption, failing to agree with measurements even qualitatively. In this study, an explicit, hybrid quantum mechanical/molecular mechanical approach for computing solvation free energies of adsorption is developed, solvation free energies of phenol adsorption are computed, and experimental data for solvation free energies of phenol adsorption are reanalyzed using multiple adsorption isotherm models. Explicit solvation calculations predict an endergonic solvation free energy for phenol adsorption that agrees well with measurements to within the experimental and force field uncertainties. Computed adsorption free energies of solvation of carbon monoxide, ethylene glycol, benzene, and phenol over the (111) facet of Pt and Cu suggest that liquid water destabilizes all adsorbed species, with the largest impact on the largest adsorbates.

Accurate Prediction of Aqueous Free Solvation Energies Using 3D Atomic Feature-Based Graph Neural Network with Transfer Learning.

Graph neural network (GNN)-based deep learning (DL) models have been widely implemented to predict the experimental aqueous solvation free energy, while its prediction accuracy has reached a plateau partly due to the scarcity of available experimental data. In order to tackle this challenge, we first build a large and diverse calculated data set Frag20-Aqsol-100K of aqueous solvation free energy with reasonable computational cost and accuracy via electronic structure calculations with continuum solvent models. Then, we develop a novel 3D atomic feature-based GNN model with the principal neighborhood aggregation (PNAConv) and demonstrate that 3D atomic features obtained from molecular mechanics-optimized geometries can significantly improve the learning power of GNN models in predicting calculated solvation free energies. Finally, we employ a transfer learning strategy by pre-training our DL model on Frag20-Aqsol-100K and fine-tuning it on the small experimental data set, and the fine-tuned model A3D-PNAConv-FT achieves the state-of-the-art prediction on the FreeSolv data set with a root-mean-squared error of 0.719 kcal/mol and a mean-absolute error of 0.417 kcal/mol using random data splits. These results indicate that integrating molecular modeling and DL would be a promising strategy to develop robust prediction models in molecular science. The source code and data are accessible at: https://yzhang.hpc.nyu.edu/IMA.

Graphical Gaussian process regression model for aqueous solvation free energy prediction of organic molecules in redox flow batteries.

The solvation free energy of organic molecules is a critical parameter in determining emergent properties such as solubility, liquid-phase equilibrium constants, pKa and redox potentials in an organic redox flow battery. In this work, we present a machine learning (ML) model that can learn and predict the aqueous solvation free energy of an organic molecule using the Gaussian process regression method based on a new molecular graph kernel. To investigate the performance of the ML model for electrostatic interaction, the nonpolar interaction contribution of the solvent and the conformational entropy of the solute in the solvation free energy, three data sets with implicit or explicit water solvent models, and contribution of the conformational entropy of the solute are tested. We demonstrate that our ML model can predict the solvation free energy of molecules at chemical accuracy with a mean absolute error of less than 1 kcal mol-1 for subsets of the QM9 dataset and the Freesolv database. To solve the general data scarcity problem for a graph-based ML model, we propose a dimension reduction algorithm based on the distance between molecular graphs, which can be used to examine the diversity of the molecular data set. It provides a promising way to build a minimum training set to improve prediction for certain test sets where the space of molecular structures is predetermined.

Prediction of aqueous free energies of solvation using coupled QM and MM explicit solvent simulations.

A method based on molecular dynamics simulations which employ two distinct levels of theory is proposed and tested for the prediction of Gibbs free energies of solvation for non-ionic solutes in water. The method consists of two additive contributions: (i) an evaluation of the free energy of solvation predicted by a computationally efficient molecular mechanics (MM) method; and (ii) an evaluation of the free energy difference between the potential energy surface of the MM method and that of a more computationally intensive first-principles quantum-mechanical (QM) method. The latter is computed by a thermodynamic integration method based on a series of shorter molecular dynamics simulations that employ weighted averages of the QM and MM force evaluations. The combined computational approach is tested against the experimental free energies of aqueous solvation for four solutes. For solute-solvent interactions that are found to be described qualitatively well by the MM method, the QM correction makes a modest improvement in the predicted free energy of aqueous solvation. However, for solutes that are found to not be adequately described by the MM method, the QM correction does not improve agreement with experiment. These preliminary results provide valuable insights into the novel concept of implementing thermodynamic integration between two model chemistries, suggesting that it is possible to use QM methods to improve upon the MM predictions of free energies of aqueous solvation.

Improving Performance of the SMD Solvation Model: Bondi Radii Improve Predicted Aqueous Solvation Free Energies of Ions and pKa Values of Thiols

Calculation of the solvation free energy of ionic molecules is the principal source of errors in the quantum chemical evaluation of pKa values using implicit polarizable continuum solvent models. One of the important parameters affecting the performance of these models is the choice of atomic radii. Here, we assess the performance of the solvation model based on density (SMD) implicit solvation model employing SMD default radii (SMD) and Bondi radii (SMD-B), a set of empirical atomic radii developed based on the crystallographic data. For a set of 112 ions (60 anions and 52 cations), the SMD-B model showed lower mean unsigned error (MUE) for predicted aqueous solvation free energies (4.0 kcal/mol for anions and 2.4 kcal/mol for cations) compared to the standard SMD model (MUE of 5.0 kcal/mol for anions and 2.9 kcal/mol for cations). In particular, usage of Bondi radii improves the aqueous solvation energies of sulfur-containing ions by >5 kcal/mol compared to the SMD default radii. Indeed, for a set of 45 thiols, the SMD-B model was found to dramatically improve the predicted pKa values, with ∼1 pKa unit mean deviation from the experimental values, compared to ∼7 pKa units mean deviation for the SMD model with the default radii. These findings highlight the importance of the choice of atomic radii on the performance of the implicit solvation models.

Energetics of CO2- in Aqueous Solution.

Gas-phase and aqueous solution properties of neutral and anionic clusters of CO2 with 3, 4, and 8 explicit H2O molecules are calculated at the coupled cluster (CCSD(T)) level plus a self-consistent reaction field. Anionic clusters with the radical electron density localized on the carbon of the CO2 molecule rather than localized on the H2O molecules are more favorable energetically by 10-20 kcal/mol in the gas phase (Δ Hgas(298 K)) and 20-30 kcal/mol in aqueous solution (Δ Gaq(298 K)). The most favorable structures are those with the largest number of strong hydrogen bonds between the CO2- and the explicit H2O molecules. Adiabatic electron affinities were calculated in the gas phase and in aqueous solution for the microsolvated anion. The adiabatic electron affinity of aqueous CO2- is predicted to be 2.35 ± 0.08 eV and is converged with as few as 3 explicit H2O molecules plus a self-consistent reaction field. The EA of aqueous CO2 is significantly greater than the aqueous solvation free energy of the electron. The vertical attachment energies to CO2 and the vertical detachment energies from CO2- were calculated. The solvated CO2- anion is substantially bent to 135°, which requires 1.52 eV. The large energy required for bending in combination with the vertical detachment and attachment energies shows that substantial local solvent reorganization occurs on detachment or attachment of an electron to solvated CO2. The formation of aqueous C2O42- from CO2- was also explored, and dimerization is predicted to occur.

Approaches for calculating solvation free energies and enthalpies demonstrated with an update of the FreeSolv database.

Solvation free energies can now be calculated precisely from molecular simulations, providing a valuable test of the energy functions underlying these simulations. Here, we briefly review "alchemical" approaches for calculating the solvation free energies of small, neutral organic molecules from molecular simulations, and illustrate by applying them to calculate aqueous solvation free energies (hydration free energies). These approaches use a non-physical pathway to compute free energy differences from a simulation or set of simulations and appear to be a particularly robust and general-purpose approach for this task. We also present an update (version 0.5) to our FreeSolv database of experimental and calculated hydration free energies of neutral compounds and provide input files in formats for several simulation packages. This revision to FreeSolv provides calculated values generated with a single protocol and software version, rather than the heterogeneous protocols used in the prior version of the database. We also further update the database to provide calculated enthalpies and entropies of hydration and some experimental enthalpies and entropies, as well as electrostatic and nonpolar components of solvation free energies.

Prediction of Mass Spectral Response Factors from Predicted Chemometric Data for Druglike Molecules.

A method is developed for the prediction of mass spectral ion counts of drug-like molecules using in silico calculated chemometric data. Various chemometric data, including polar and molecular surface areas, aqueous solvation free energies, and gas-phase and aqueous proton affinities were computed, and a statistically significant relationship between measured mass spectral ion counts and the combination of aqueous proton affinity and total molecular surface area was identified. In particular, through multilinear regression of ion counts on predicted chemometric data, we find that log10(MS ion counts) = -4.824 + c 1•PA + c 2•SA, where PA is the aqueous proton affinity of the molecule computed at the SMD(aq)/M06-L/MIDI!//M06-L/MIDI! level of electronic structure theory, SA is the total surface area of the molecule in its conjugate base form, and c 1 and c 2 have values of -3.912 × 10-2 mol kcal-1 and 3.682 × 10-3 Å-2. On a 66-molecule training set, this regression exhibits a multiple R value of 0.791 with p values for the intercept, c 1, and c 2 of 1.4 × 10-3, 4.3 × 10-10, and 2.5 × 10-6, respectively. Application of this regression to an 11-molecule test set provides a good correlation of prediction with experiment (R = 0.905) albeit with a systematic underestimation of about 0.2 log units. This method may prove useful for semiquantitative analysis of drug metabolites for which MS response factors or authentic standards are not readily available. Graphical Abstract ᅟ.

KECSA-Movable Type Implicit Solvation Model (KMTISM).

Computation of the solvation free energy for chemical and biological processes has long been of significant interest. The key challenges to effective solvation modeling center on the choice of potential function and configurational sampling. Herein, an energy sampling approach termed the “Movable Type” (MT) method, and a statistical energy function for solvation modeling, “Knowledge-based and Empirical Combined Scoring Algorithm” (KECSA) are developed and utilized to create an implicit solvation model: KECSA-Movable Type Implicit Solvation Model (KMTISM) suitable for the study of chemical and biological systems. KMTISM is an implicit solvation model, but the MT method performs energy sampling at the atom pairwise level. For a specific molecular system, the MT method collects energies from prebuilt databases for the requisite atom pairs at all relevant distance ranges, which by its very construction encodes all possible molecular configurations simultaneously. Unlike traditional statistical energy functions, KECSA converts structural statistical information into categorized atom pairwise interaction energies as a function of the radial distance instead of a mean force energy function. Within the implicit solvent model approximation, aqueous solvation free energies are then obtained from the NVT ensemble partition function generated by the MT method. Validation is performed against several subsets selected from the Minnesota Solvation Database v2012. Results are compared with several solvation free energy calculation methods, including a one-to-one comparison against two commonly used classical implicit solvation models: MM-GBSA and MM-PBSA. Comparison against a quantum mechanics based polarizable continuum model is also discussed (Cramer and Truhlar’s Solvation Model 12).

Composite method for implicit representation of solvent in dimethyl sulfoxide and acetonitrile.

A composite method for implicit representation of solvent previously developed to compute aqueous free energies of solvation is extended to accommodate the polar aprotic solvents dimethyl sulfoxide and acetonitrile. The method combines quantum mechanical calculation of the solute electronic structure with a modern dielectric continuum model for long-range electrostatic interactions with solvent and individual models for short-range interactions arising from dispersion, exchange, and hydrogen bonding. The few parameters involved are optimized to fit a standard data set of experimental solvation energies for neutrals and ions. Results are better than other models in the literature, with average errors for ions comparable to or smaller than the estimated experimental errors. Some circumstantial evidence is also obtained to support one of the competing extrathermodynamic arguments recently used to determine the solvation energies of the proton, which are needed to separate measurements of paired cation plus anion solvation energies into absolute single ion solvation energies in these solvents.

Molecular Dynamics of β-Hairpin Models of Epigenetic Recognition Motifs

The conformations and stabilities of the β-hairpin model peptides of Waters (Riemen, A. J.; Waters, M. L. Biochemistry 2009, 48, 1525; Hughes, R. M.; Benshoff, M. L.; Waters, M. L. Chemistry 2007, 13, 5753) have been experimentally characterized as a function of lysine ε-methylation. These models were developed to explore molecular recognition of known epigenetic recognition motifs. This system offered an opportunity to computationally examine the role of cation-π interactions, desolvation of the ε-methylated ammonium groups, and aromatic/aromatic interactions on the observed differences in NMR spectra. AMOEBA, a second-generation force field (Ponder, J. W.; Wu, C.; Ren, P.; Pande, V. S.; Chodera, J. D.; Schnieders, M. J.; Haque, I.; Mobley, D. L.; Lambrecht, D. S.; DiStasio, R. A., Jr.; Head-Gordon, M.; Clark, G. N.; Johnson, M. E.; Head-Gordon, T. J. Phys. Chem. B 2010, 114, 2549), was chosen as it includes both multipole electrostatics and polarizability thought to be essential to accurately characterize such interactions. Independent parametrization of ε-methylated amines was required from which aqueous solvation free energies were estimated and shown to agree with literature values. Molecular dynamics simulations (100 ns) using the derived parameters with model peptides, such as Ac-R-W-V-W-V-N-G-Orn-K(Me)(n)-I-L-Q-NH(2), where n = 0, 1, 2, or 3, were conducted in explicit solvent. Distances between the centers of the indole rings of the two-tryptophan residues, 2 and 4, and the ε-methylated ammonium group on Lys-9 as well as the distance between the N- and C-termini were monitored to estimate the strength and orientation of the cation-π and aromatic/aromatic interactions. In agreement with the experimental data, the stability of the β-hairpin increased significantly with lysine ε-methylation. The ability of MD simulations to reproduce the observed NOEs for the four peptides was further estimated for the monopole-based force fields, AMBER, CHARMM, and OPLSAA. AMOEBA correctly predicted over 80% of the observed NOEs for all 4 peptides, while the three-monopole force fields were 40-50% predictive in only 2 cases and approximately 10% in the other 10 examples. Preliminary analysis suggests that the decreased cost of desolvation of the substituted ammonium group significantly compensated for the reduced cation-π interaction resulting from the increased separation due to steric bulk of the ε-methylated amines.

Conformational Analysis of 6α- and 6β-Naltrexol and Derivatives and Relationship to Opioid Receptor Affinity

Naltrexol and its C₆ α and β desoxy, iodo, mesyl, tosyl, trifyl, dimethylcarbamyl, and diphenylcarbamyl derivatives were studied in their energy-minimized C ring chair-like and boat-like conformations using B3LYP/6-31G** and SM5.4/A to estimate aqueous solvation free energy. The results were compared to experimental opioid receptor binding affinities. The total energy difference between β conformers correlated well with MOR binding affinity, while the aqueous solvation free energy correlated well with the KOR binding affinity.

Prediction of SAMPL2 aqueous solvation free energies and tautomeric ratios using the SM8, SM8AD, and SMD solvation models

We applied the solvation models SM8, SM8AD, and SMD in combination with the Minnesota M06-2X density functional to predict vacuum-water transfer free energies (Task 1) and tautomeric ratios in aqueous solution (Task 2) for the SAMPL2 test set. The bulk-electrostatic contribution to the free energy of solvation is treated as follows: SM8 employs the generalized Born model with the Coulomb field approximation, SM8AD employs the generalized Born approximation with asymmetric descreening, and SMD solves the nonhomogeneous Poisson equation. The non-bulk-electrostatic contribution arising from short-range interactions between the solute and solvent molecules in the first solvation shell is treated as a sum of terms that are products of geometry-dependent atomic surface tensions and solvent-accessible surface areas of the individual atoms of the solute. On average, three models tested in the present work perform similarly. In particular, we achieved mean unsigned errors of 1.3 (SM8), 2.0 (SM8AD), and 2.6 kcal/mol (SMD) for the aqueous free energies of 30 out of 31 compounds with known reference data involved in Task 1 and mean unsigned errors of 2.7 (SM8), 1.8 (SM8AD), and 2.4 kcal/mol (SMD) in the free energy differences (tautomeric ratios) for 21 tautomeric pairs in aqueous solution involved in Task 2.

Free Energies of Solvation with Surface, Volume, and Local Electrostatic Effects and Atomic Surface Tensions to Represent the First Solvation Shell.

Building on the SVPE (surface and volume polarization for electrostatics) model for electrostatic contributions to the free energy of solvation with explicit consideration of both surface and volume polarization effects, on the SMx approach to including first-solvation-shell contributions, and on the linear relationship between the electric field and short-range electrostatic contributions found by Chipman, we have developed a new method for computing absolute aqueous solvation free energies by combining the SVPE method with semiempirical terms that account for effects beyond bulk electrostatics. The new method is called SMVLE, and the elements it contains are denoted by SVPE-CDSL where SVPE denotes accounting for bulk electrostatic interactions between solute and solvent with both surface and volume contributions, CDS denotes the inclusion of solvent cavitation, changes in dispersion energy, and possible changes in local solvent structure by a semiempirical term utilizing geometry-dependent atomic surface tensions as implemented in SMx models, and L represents the local electrostatic effect derived from the outward-directed normal electric field on the cavity surface. The semiempirical CDS and L terms together represent the deviation of short-range contributions to the free energy of solvation from those accounted for by the SVPE term based on the bulk solvent dielectric constant. A solute training set containing a broad range of molecules used previously in the development of SM6 is used here for SMVLE model calibration. The aqueous solvation free energies predicted by the parameterized SMVLE model correlate exceedingly well with experimental values. The square of the correlation coefficient is 0.9949 and the slope is 1.0079. Comparison of the final SMVLE model against the earlier SMx solvation model shows that the parameterized SMVLE model not only yields good accuracy for neutrals but also significantly increases the accuracy for ions, making it the best implicit solvation model to date for aqueous solvation free energies of ions. The semiempirical terms associated with the outward-directed electric field account in a physical way for the improvement in the predictive accuracy for ions. The SMVLE method greatly decreases the need to include explicit water molecules for accurate modeling of solvation free energies of ions.

Universal Solvation Model Based on Solute Electron Density and on a Continuum Model of the Solvent Defined by the Bulk Dielectric Constant and Atomic Surface Tensions

We present a new continuum solvation model based on the quantum mechanical charge density of a solute molecule interacting with a continuum description of the solvent. The model is called SMD, where the "D" stands for "density" to denote that the full solute electron density is used without defining partial atomic charges. "Continuum" denotes that the solvent is not represented explicitly but rather as a dielectric medium with surface tension at the solute-solvent boundary. SMD is a universal solvation model, where "universal" denotes its applicability to any charged or uncharged solute in any solvent or liquid medium for which a few key descriptors are known (in particular, dielectric constant, refractive index, bulk surface tension, and acidity and basicity parameters). The model separates the observable solvation free energy into two main components. The first component is the bulk electrostatic contribution arising from a self-consistent reaction field treatment that involves the solution of the nonhomogeneous Poisson equation for electrostatics in terms of the integral-equation-formalism polarizable continuum model (IEF-PCM). The cavities for the bulk electrostatic calculation are defined by superpositions of nuclear-centered spheres. The second component is called the cavity-dispersion-solvent-structure term and is the contribution arising from short-range interactions between the solute and solvent molecules in the first solvation shell. This contribution is a sum of terms that are proportional (with geometry-dependent proportionality constants called atomic surface tensions) to the solvent-accessible surface areas of the individual atoms of the solute. The SMD model has been parametrized with a training set of 2821 solvation data including 112 aqueous ionic solvation free energies, 220 solvation free energies for 166 ions in acetonitrile, methanol, and dimethyl sulfoxide, 2346 solvation free energies for 318 neutral solutes in 91 solvents (90 nonaqueous organic solvents and water), and 143 transfer free energies for 93 neutral solutes between water and 15 organic solvents. The elements present in the solutes are H, C, N, O, F, Si, P, S, Cl, and Br. The SMD model employs a single set of parameters (intrinsic atomic Coulomb radii and atomic surface tension coefficients) optimized over six electronic structure methods: M05-2X/MIDI!6D, M05-2X/6-31G, M05-2X/6-31+G, M05-2X/cc-pVTZ, B3LYP/6-31G, and HF/6-31G. Although the SMD model has been parametrized using the IEF-PCM protocol for bulk electrostatics, it may also be employed with other algorithms for solving the nonhomogeneous Poisson equation for continuum solvation calculations in which the solute is represented by its electron density in real space. This includes, for example, the conductor-like screening algorithm. With the 6-31G basis set, the SMD model achieves mean unsigned errors of 0.6-1.0 kcal/mol in the solvation free energies of tested neutrals and mean unsigned errors of 4 kcal/mol on average for ions with either Gaussian03 or GAMESS.

Predictions of hydration free energies from all-atom molecular dynamics simulations.

Here, we computed the aqueous solvation (hydration) free energies of 52 small drug-like molecules using an all-atom force field in explicit water. This differs from previous studies in that (1) this was a blind test (in an event called SAMPL sponsored by OpenEye Software) and (2) the test compounds were considerably more challenging than have been used in the past in typical solvation tests of all-atom models. Overall, we found good correlations with experimental values which were subsequently made available, but the variances are large compared to those in previous tests. We tested several different charge models and found that several standard charge models performed relatively well. We found that hypervalent sulfur and phosphorus compounds are not well handled using current force field parameters and suggest several other possible systematic errors. Overall, blind tests like these appear to provide significant opportunities for improving force fields and solvent models.

Absolute and relative p Ka calculations of mono and diprotic pyridines by quantum methods

Gas-phase deprotonation free energies and aqueous solvation free energies of 10 monoprotic and 6 diprotic pyridines were calculated using CBS-QB3 and HF/CPCM or B3LYP/CPCM methods. The results have been processed considering three thermodynamic cycles for absolute and relative p K a calculations. In these cycles, different experimental values of solvation free energies for proton, water and hydronium were considered. Moreover, calculated ab initio and DFT values were used when possible. Results show that the inclusion of explicit single water molecule interacting with pyridine nitrogen improves predictions in 1.5 p K a units. Water molecule causes an increase in the solute–solvent surface interaction and allows the continuum method to reproduce correct solvation free energy differences between acids and bases. The correct combination of computational methodology and thermodynamic cycle leads to very accurate results, with mean absolute errors of 0.3–0.5 p K a units for monoprotic and 0.7–0.9 p K a units for diprotic pyridines.

Modeling molecular and ionic absolute solvation free energies with quasichemical theory bounds

A recently developed statistical mechanical quasichemical theory (QCT) has led to significant insights into solvation phenomena for both hydrophilic and hydrophobic solutes. The QCT exactly partitions solvation free energies into three components: (1) Inner-shell chemical, (2) outer-shell packing, and (3) outer-shell long-ranged contributions. In this paper, we discuss efficient methods for computing each of the three parts of the free energy. A Bayesian estimation approach is developed to compute the inner-shell chemical and outer-shell packing contributions. We derive upper and lower bounds on the outer-shell long-ranged portion of the free energy by expressing this component in two equivalent ways. Local, high-energy contacts between the solute and solvent are eliminated by spatial conditioning in this free energy piece, leading to near-Gaussian distributions of solute-solvent interaction energies. Thus, the average of the two mean-field bounds yields an accurate and efficient free energy estimate. Aqueous solvation free energy results are presented for several solutes, including methane, perfluoromethane, water, and sodium and chloride ions. The results demonstrate the accuracy and efficiency of the methods. The approach should prove useful in computing solvation free energies in inhomogeneous, restricted environments.

Prediction of the aqueous solvation free energy of organic compounds by using autocorrelation of molecular electrostatic potential surface properties combined with response surface analysis

Several quantitative structure–property relationship (QSPR) approaches have been explored for the prediction of aqueous solubility or aqueous solvation free energies, Δ G sol, as crucial parameter affecting the pharmacokinetic profile and toxicity of chemical compounds. It is mostly accepted that aqueous solvation free energies can be expressed quantitatively in terms of properties of the molecular surface electrostatic potentials of the solutes. In the present study we have introduced autocorrelation molecular electrostatic potential ( autoMEP) vectors in combination with nonlinear response surface analysis (RSA) as alternative 3D-QSPR strategy to evaluate the aqueous solvation free energy of organic compounds. A robust QSPR model ( r cv = 0.93) has been obtained by using a collection of 248 organic chemicals. An external test set based on 23 molecules confirmed the good predictivity of the autoMEP/RSA model suggesting its further applicability in the in silico prediction of water solubility of large organic compound libraries.