Prediction of boiling points of organic compounds by QSPR tools

Spectral deep learning for prediction and prospective validation of functional groups

“…The arrangement of atoms within a molecule dictates its physical, chemical, and spectral properties. Small discrete, or large repeating arrangements of atoms which give rise to measurable changes in a molecule's reactivity, [1][2][3] boiling point, 4,5 melting point, 6,7 and other characteristics are called functional groups. Given the structural formula of a molecule, a chemist can identify functional groups present (e.g.…”

A Systematic Method for Selecting Molecular Descriptors as Features When Training Models for Predicting Physiochemical Properties

“…In general, smaller or molecular family-specific (e.g., alcohols, alkanes, hydrocarbons, unsaturated hydrocarbons, and heterogeneous molecules) data sets result in lower errors for the boiling point, flash point, and yield sooting index models [11,12,16,17]. Melting point and heat of combustion models, however, have lower errors with the largest data sets [10,22].…”

“…Some studies only reported training or overall errors that averaged training, testing, and validation (when available) errors [1,2,12,15,39]. As noted by other researchers, reporting the test error provides a better measure of the model's predictive capability because external validation is necessary for determining the true predictive ability of the model [17].…”

“…Preliminary fuel screening tools based on machine learning have already proven useful in programs worldwide and help identify promising molecules early in the fuel development cycle [1][2][3][4][5][6]. Many of these property prediction models also use chemical descriptors as features to provide a mathematical link between physiochemical properties and molecular structure [1][2][3][7][8][9][10][11][12][13][14][15][16][17][18][19][20][21][22][23][24] features in quantitative structure-property relationship models can provide new information about the property predicted, and lead to better understanding of the link between physical or chemical properties and molecular structure [32]. Many predictive models developed for identifying correlations between descriptors and properties of interest (i.e., interpretability) use multi-linear equations or least-squares regression [3,[7][8][9]15,[17][18][19][20][21].…”

Modelling of Normal Boiling Points of Hydroxyl Compounds by Radial Basis Networks

“…The method is applicable only to the compounds for which all group contributions have been established. Another well-known solution is quantitative structure-property relationships (QSPR) approach [7][8][9]. In this approach, a QSPR model is introduced by developing a correlation between the NBP and a variety of molecular features.…”