TY - JOUR
T1 - The role of the dominant descriptor in targeted quantitative structure property relationships
AU - Shacham, Mordechai
AU - Kahrs, Olaf
AU - Cholakov, Georgi St
AU - Stateva, Roumiana P.
AU - Marquardt, Wolfgang
AU - Brauner, Neima
PY - 2007/11
Y1 - 2007/11
N2 - The targeted quantitative structure-property relationship (TQSPR) method of Brauner et al. [2006. A structurally "targeted" QSPR method for property prediction. Industrial & Engineering Chemistry Research 45, 8430-8437] is analyzed in this study with examples involving prediction of critical temperatures of hydrocarbons. Targeted are 108 hydrocarbons, for which experimental critical temperature values can be obtained from published sources. Each hydrocarbon structure is represented by 1630 calculated values of molecular descriptors. The availability of a wide variety of molecular descriptors and small training sets comprising compounds structurally similar to the target enables the detection of one, dominant descriptor, which is well correlated with a particular property and is sufficient to serve as a single predictor for this property. The analysis of the relationship between the dominant descriptor and property values for the training set and the target compound enables better understanding of the main causes for increased prediction errors. Based on this analysis, new quantitative indicators for assessing the reliability of the predicted property value are proposed, which may reduce considerably the uncertainty associated with property prediction.
AB - The targeted quantitative structure-property relationship (TQSPR) method of Brauner et al. [2006. A structurally "targeted" QSPR method for property prediction. Industrial & Engineering Chemistry Research 45, 8430-8437] is analyzed in this study with examples involving prediction of critical temperatures of hydrocarbons. Targeted are 108 hydrocarbons, for which experimental critical temperature values can be obtained from published sources. Each hydrocarbon structure is represented by 1630 calculated values of molecular descriptors. The availability of a wide variety of molecular descriptors and small training sets comprising compounds structurally similar to the target enables the detection of one, dominant descriptor, which is well correlated with a particular property and is sufficient to serve as a single predictor for this property. The analysis of the relationship between the dominant descriptor and property values for the training set and the target compound enables better understanding of the main causes for increased prediction errors. Based on this analysis, new quantitative indicators for assessing the reliability of the predicted property value are proposed, which may reduce considerably the uncertainty associated with property prediction.
KW - Computational chemistry
KW - Parameter identification
KW - Process design
KW - Product design
KW - Property prediction
KW - QSPR
UR - http://www.scopus.com/inward/record.url?scp=34848922721&partnerID=8YFLogxK
U2 - 10.1016/j.ces.2007.07.029
DO - 10.1016/j.ces.2007.07.029
M3 - ???researchoutput.researchoutputtypes.contributiontojournal.article???
AN - SCOPUS:34848922721
SN - 0009-2509
VL - 62
SP - 6222
EP - 6233
JO - Chemical Engineering Science
JF - Chemical Engineering Science
IS - 22
ER -