Quantitative Structure–property Relationshipanalysis of Critical Properties of Individual Organic Compounds

This work presents the development and analysis of robust QSPR models for critical properties (pressure PC, temperature TC, and volume VC). The object of this study is a database consisting of 399 different organic compounds. These compounds include saturated and unsaturated hydrocarbons, aromatic hydrocarbons, heterocyclic compounds, alcohols, simple and complex ethers, and their derivatives. The database encompasses carbon-, halogen-, oxygen-, nitrogen-, and sulfur- containing compounds. The molecular structures of investigated compounds were previously modeled, standardized, and validated with respect to connectivity and uniqueness. Structural descriptors were calculated at the 2D molecular modeling level using the simplex approach. To differentiate simplex vertices, not only atom types but also other atomic characteristics were used, including van der Waals interactions (from the universal force field), and potentials of informational fields weighted by atomic properties such as partial charge, polarizability, electronegativity, and lipophilicity. In addition to simplexes, smaller fragments consisting of two and three atoms were also used for each compound of database. A total of 4,939 2D structural descriptors were calculated. The Random Forest (RF) method was applied to establish the relationships between the structural descriptors and the critical properties (PC, TC, VC). The developed 2D RF QSPR models showed high approximation accuracy (R²=0.99) and predictive ability (R²oob=0.90...0.97). Physicochemical interpretation revealed that electrostatic factors have the greatest impact on critical properties. For compounds lacking experimental data on the studied properties, predictions were presented using the developed RF models, and the applicability domain of the developed QSPR models was evaluated.