MODELING AND PREDICTION OF AMINO ACIDS LIPOPHYLICITY USING MULTIPLE LINEAR REGRESSION COUPLED WITH GENETIC ALGORITHM
Alexandrina Guidea, Costel Sârbu · Studia Universitatis Babeș-Bolyai Chemia · 2019
Quantitative structure-retention relationships (QSRR) approach was used to model chromatographic lipophilicity of sixteen proteinogenic amino acids using molecular descriptors computed with DRAGON and ALCHEMY software packages. Modeling was performed applying multiple linear regression (MLR) coupled with genetic algorithms (GA) methodology (MLR-GA). The most important descriptors, highly significant in the predictive models of amino acids lipophilicity (RM0), were related to atomic polarizabilities (MATS3p; Ap; H1p), atomic van der Waals volume (MATS3v), Sanderson electronegativity (RDF070e) and Randic molecular profiles (DP11; DP12) calculated with Dragon software. The internal statistical evaluation procedure highlighted some appropriate models for the chromatographic lipophilicity prediction. Moreover, the statistical parameters of regression in order to evaluate the relationship between experimental and predicted values, in case of the test set (four amino acids), revealed three statistically valid models (model A, E and F) that can be successfully used in lipophilicity prediction of amino acids.