engleski

Modeling blood-brain barrier penetration using descriptors computed by the DRAGON and CERIUS2

Relationship between molecular structure encoded by molecular descriptors and blood- brain barrier penetration of 106 organic compounds were studied using an algorithm for selection of the most important subset of descriptors in the multivariate regression models. The blood-brain barrier (BBB) penetration is measured experimentally as the ratio of the concentration of the compound in the brain, to that in the blood (in modeling, logarithm of BBB values was used). Initial structures of molecules were encoded as SMILES and converted to the 3D structures by the CORINA program (www2.chemie.uni- erlangen.de/software/corina/) and more than 800 initial descriptors were computed by the program Dragon 4.0 (http://www.disat.unimib.it/chm/), and more than 300 descriptors were computed by the Cerius2 program (Molecular Simulations Inc., San Diego). Descriptors from certain classes, which are not so easy to interpret, as well as various computed logP descriptors were omitted. The initial set of descriptors was filtered in order to remove nonsignificant and highly intercorrelated descriptors - 330 (Dragon) and 160 (Cerius2) descriptors remained after filtering. Finally, the best linear multiregression models containing three and four descriptors were selected by the CROMRsel program [1]. Efficiency of this program for selecting the small set of the most relevant molecular descriptors into the linear and nonlinear multivariate regression models was confirmed in several comparative studies (see for example refs. [1-3]). The standard error of estimate for the best 3-descriptor model (descriptors computed by Dragon 4.0, four outliers are removed) is 0.422 log units. Descriptors involved in the model are the sum of atomic Sanderson electronegativities, the quadratic index (topological descriptor), and the total information content index (neighborhood symmetry of 1-order). The standard error estimate for the best 4- descriptor model (descriptors computed by Cerius2, four outliers are removed) is 0.450 log units. Descriptors involved in the model are the desolvation free energy for water, the number of rotatable bonds, the atom type index, and the molar refractivity. This result favors the Dragon set of descriptors over the one computed by the Cerius2 program. Models obtained on both sets of descriptors are additionally improved by the inclusion of nonlinear terms of initial descriptors. These models are slightly better than the best available models in the literature for the same set of molecules [4, 5]. [1] Lučić B, Trinajstić N. Multivariate regression outperforms several robust architectures of neural networks in QSAR modeling. J Chem Inf Comput Sci 1999 ; 39:121-132. [2] Lučić B, Amić D, Trinajstić N. Nonlinear multivariate regression outperforms several concisely designed neural networks on three QSPR data sets. J Chem Inf Comput Sci 2000 ; 40:403-413. [3] Lučić B, Nadramija D, Bašic I, Trinajstić N. Toward generating simpler QSAR models: nonlinear multivariate regression versus several neural network ensembles and some related methods. J Chem Inf Comput Sci 2003 ; 43:1094-1102. [4] Rose K, Hall LH, Kier LB. Modeling blood-brain barrier partitioning using the electrotopological state. J Chem Inf Comput Sci 2002 ; 42:651-666. [5] Rose K, Hall LH, Hall LM, Kier LB. Modeling blood-brain barrier partitioning using topological structure descriptors. 1-13, in MDL® ; ; Discovery Predictive Science, San Leandro (CA): MDL Information Systems, Inc. ; 2003. Available from: URL: www.mdl.com

blood-brain barrier penetration ; molecular structure ; SMILES structure ; CORINA program ; Dragon 4.0 ; Cerius2 ; modeling ; selecton of descriptors ; CROMRsel program ; multivariate regresion

nije evidentirano