Topological Index Computing System And Its Applications

It is key problem that how to convert numerical value from molecular structure (2D/3D) and that how to build good model in quantitive structure-property/activity relationship (QSPR/QSAR) study. Each molecular structure can be represented through different approach such as topological index, electronic parameter, geometric parameter and quantum chemistry parameter, but topological index due to its simple calculation is widely used in QSPR/QSAR. So far, much more than one hundred topological indices have been proposed and used to correlate with the properties or activities of compounds. The final purpose of model is able to predict the property/activity of testing compounds, so the methods of viable selection and building model are important in process of building model.The main contents and some conclusions are as follows:1. In this paper, molecular 3D structure files which were saved in computer are converted to HQN-code. The calculation system of topological index, TOPO.1, is programmed, which can calculate some important topological indices such as molecular connective index, electrotopological state index, Kappa shape index and molecular electronegativity-distance vector, and can also carry some statistical analysis.2. An algorithm is proposed to detect the aromaticity of molecular structures based on HQN connection matrix acquired by HQN-code. The blindness of the new algorithm is reduced, to the most extent. Examples show that the computer program can identify the aromatic structure, and the result of identification are concise and effective. The algorithm has been implemented in software TOPO. 13. GAPLS is a variable selection method combining with Genetic Algorithms(GA), back stepwise and partial least squares (PLS), which is introduced in the variable selection of quantitative structure GC retention index relationship. Seven topological indices are selected from 127topological indices by GAPLS to build QSRR model. The leave-one-out cross-validation indicates the built model is reliable and stable. The seven variables are also interpreted structurally.4. The model of the brain-blood distribution of a large and diverse set of organic drug molecules, using the method of orthogonal block variables and canonical correlation based on topological descriptors, is much better than before. This model, with correlation coefficient 0.9927 and standard deviation 0.0946, has not only a good fitting feature but also a very good prediction ability, with correlation coefficient of predictions 0.992 and standard deviation of predictions 0.0999, respectively, through leave-one-out cross validation analysis. It might be of general utility in predicting partition coefficient for a very wide range of new drugs.