Premium
Property prediction by correlations based on similarity of molecular structures
Author(s) -
Shacham Mordechai,
Brauner Neima,
Cholakov Georgi,
Stateva Roumiana P.
Publication year - 2004
Publication title -
aiche journal
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.958
H-Index - 167
eISSN - 1547-5905
pISSN - 0001-1541
DOI - 10.1002/aic.10248
Subject(s) - molecular descriptor , consistency (knowledge bases) , similarity (geometry) , property (philosophy) , range (aeronautics) , regression , biological system , quantitative structure–activity relationship , correlation , mathematics , chemical similarity , data mining , linear regression , computer science , statistics , artificial intelligence , machine learning , materials science , philosophy , geometry , epistemology , composite material , image (mathematics) , biology , cluster analysis
A new approach for predicting a wide range of physical and thermodynamic properties is proposed. It involves calculation of the molecular descriptors of a target compound of unknown properties, followed by regression of this vector of molecular descriptors vs. a database of compounds with known descriptors and measured properties. The regression model, obtained for the target descriptors in terms of predictive compounds and their coefficients, is then used for prediction of properties of the target compound. The precision of the prediction can be estimated from the standard deviation of the correlation and the known precision of the property data of the predictive compounds. The proposed method was tested in predicting 31 properties of 18 compounds representing different hydrocarbon structures. The results show that the method has several unique advantages, such as the use of one structural correlation to predict all properties; estimation of the prediction error for compounds without measured data; opportunities to find alternative solutions to different problems and means to estimate their adequacy. The method can be used also for checking the consistency of measured data and data predicted by other methods. © 2004 American Institute of Chemical Engineers AIChE J, 50: 2481–2492, 2004