Study On Prediction Of Biology Activity Based On Molecular Indices

An early prediction of the in-vivo ADME/T especially the absorption and metabolismproperties at the early stage of drug discovery is very important for reducing the the risk ofinvesting by a long way and decreasing the development time largely. Human intestinalabsorption and blood brain barrier (BBB) permeation abilities are two crucial absorptioncapabitities for drugs, and the P450 metabolism enzymes are key targets on the study ofmetabilization. While presently the experimental determinations of these properties areusually very troublesome or hardly to be performed. While computer aided drug design(CADD) appears and gives direction to the drug development and discovery, this methodincreases the efficiency and successful rate of drug discovery largely, in which the 2-dimensional QSAR (2D QSAR) is widely used by its obvious advantages of fast and simplecalculations of molecular indices which are easily interpreted.In the present work, a 2D QSAR method combining the multivariate linear regressionanalysis method, principal component analysis, discriminant analysis method and K-MeansCluster method is applied to the study of several kinds of drugs, resulting in several reliablemodels. Proper statistical results are achieved for these models which are similar to or betterthan the reference. Based on calculated molecular indices, our work includes:1) The human intestinal absorption properties of 100 diverse compounds are predicted byseveral regression methods. By using the optimal one, a model with satisfactory statisticalresults (R~2＞0.80) and good predictability is established.2) The BBB permeation properties of 190 different kinds of compounds are predicted by amodel we established, which correlates well the molecular indices with BBB permeation,and those key descriptors impacting the molecules' biological properties are identified.3) A discriminant mathematical model based on 81 compounds (35 P450 2C9 inhibitors and46 non-inhibitors) is built. The statistical result of the model is that the discriminantaccuracies for inhibitors and non-inhibitors are 96.4% and 90.0% respectively. This workwill be helpful for aiding further screening and discovery of P450 2C9 inhibitors.