A General Treatment of Solubility. 3. Principal Component Analysis (PCA) of the Solubilities of Diverse Solutes in Diverse Solvents

Abstract

A phenomenological study of solubility has been conducted using a combination of quantitative structureproperty relationship (QSPR) and principal component analysis (PCA). A solubility database of 4540 experimental data points was used that utilized available experimental data into a matrix of 154 solvents times 397 solutes. Methodology in which QSPR and PCA are combined was developed to predict the missing values and to fill the data matrix. PCA on the resulting filled matrix, where solutes are observations and solvents are variables, shows 92.55% of coverage with three principal components. The corresponding transposed matrix, in which solvents are observations and solutes are variables, showed 62.96% of coverage with four principal components.

Predicting hydrophobic solvation by molecular simulation: 2. New united‐atom model for alkanes, alkenes, and alkynes

“…It was found that all three force-fields showed systematic deviations from experimental data [18,19], with OPLS-UA and GROMOS overestimating the magnitude of solvation (by 15% and 13%, respectively), and TraPPE slightly underestimating it (by 6%) [17]. This performance was rationalized on the basis of the parameterization strategy and target experimental properties used by each model.…”

“…However, it appears that within this range of values, the influence of the solute-solvent attraction dominates and the excluded volume effect is rather minor. [18,19], while the thin dashed lines represent upper and lower bounds based on the estimated uncertainty in experimental measurements [40].…”

“…Also, cyclohexane is a widely used solvent, so this system assumes particular relevance for future applications of the model. As target experimental properties, the density of the liquid [37], the enthalpy of vaporization [38] and the self-solvation free energy (i.e., for cyclohexane solute dissolved in cyclohexane solvent) [18,19] were chosen.Analyzing the parameters for cyclic CH2 groups in the 3 force-fields considered earlier (see Table 1 of the previous paper [17]), it can be seen that they are spread over a relatively narrow range of values around  ≈ 0.39 nm and  ≈ 0.46 kJ/mol. Therefore, the sensitivity of the three different target properties to  and  was probed over a narrow window roughly centered on those values.…”

“…However, it appears that within this range of values, the influence of the solute-solvent attraction dominates and the excluded volume effect is rather minor. [18,19], while the thin dashed lines represent upper and lower bounds based on the estimated uncertainty in experimental measurements [40].Also shown in Figures 1-3 are the experimental values for each property, with corresponding uncertainties (for density, this is assumed to be negligible). It is clear that for a given property there exists a potentially infinite set of (, ) pairs that can match the experimental value.…”

“…Perhaps surprisingly, it was not possible to find any UA models of alkynes in the literature (parameters exist for all-atom models, but these contain explicit hydrogens and point charges, so they are not suitable for our purposes). Therefore, a new parameter set was developed from scratch, aiming to reproduce solvation free energies and pure liquid densities of alkyne molecules.The first step was to determine parameters for CH groups with sp hybridization by matching the experimental solvation free energy of acetylene in n-heptane [18,19] and the density of pure acetylene at its standard boiling point [37]. The parameterization strategy was very similar to the one described above for methane (section 3.2), except that here one did not have a good initial guess for the parameters.…”