Use of the DIPPR Database for Development of QSPR Correlations:  Surface Tension

• Published in: Journal of chemical and engineering data 2001-09, Vol.46 (5), p.1007-1012
• Main Authors: Knotts, Thomas A., Wilding, W. Vincent, Oscarson, John L., Rowley, Richard L.
• Format: Article
• Language: English
• Subjects: Chemistry; Exact sciences and technology; General and physical chemistry; General, apparatus; General. Nomenclature, chemical documentation, computer chemistry; Surface physical chemistry; Theory of reactions, general kinetics. Catalysis. Nomenclature, chemical documentation, computer chemistry
• ISSN: 0021-9568; 1520-5134
• DOI: 10.1021/je000232d

Combination of commercial QSPR (quantitative structure−property relationship) software with an evaluated database creates a powerful tool for development of thermophysical property correlations. By using data quality codes in the DIPPR relational database, a training set of property values within a desired accuracy level can be obtained for use in QSPR regression software. Moreover, additional database queries can be used to restrict the training set to specified families or functional groups and further refine the molecular descriptors that are used to correlate the property. This provides a good basis for rapid development of QSPR correlations of known uncertainty and chemical domain. This procedure is illustrated by its application to the extension of the Macleod−Sugden (Trans. Faraday Soc. 1923, 19, 38. Chem. Soc. 1924, 125, 32.) correlation for surface tension based upon the parachor. Quayle (Chem. Rev. 1953, 53, 439−591.) correlated the parachor in terms of additive atomic and structural increments but used a training set limited in temperature and scope. In this work, new molecular descriptors were selected consistent with the accuracy of the training set extracted from the DIPPR database, and their additive increments to the parachor were regressed from 8697 surface tension values of uncertainty less than 5% for 649 different compounds. This produced a correlation with an average absolute deviation (AAD) of 3.2%. This can be compared with an AAD of 6.9% using the Quayle descriptors for the same set.