Quantitative Structure Activity Relationships Study Of Blood Brain Partitioning And Human Intestinal Absorption

The drug develop process is becoming more expensive and time-consuming. A main reason for high attrition rates in drug discovery is from the Absorption, Distribution, Metabolism and Excretion (ADME) properties of candidate compounds. So ADME studies have to be embraced in the drug discovery process, and better ADME properties are pursued. Computer models are available for predicting ADME properties with less expense and time.In the first part of this thesis, two models of Blood brain partitioning (LogBB) are built by multiple linear regression(MLR) and support vector machine(SVM) regression. A data set of 53 compounds is taken from previous literatures for training set of 41 compounds and test set of 12 compounds. Four descriptors (LogP, LogS, HD and MW) are applied to describe the structural characteristics of organic compounds and drugs. For training set, coefficients (r) are 0.88 and 0.90, standard deviations (s) are 0.31 and 0.29, and mean abstract errors (MAE) are 0.25 and 0.23. Root mean squares (RMS) are 0.28 and 0.26 for entire models.In the second part of this thesis, a dataset of 552 compounds with Human Intestinal Absorption (HIA) properties has been investigated. HIA prediction models are built. Descriptors are generated by ADRIANA.Code and Cerius~2. A Genetic Algorithm feature selection method is applied to select proper descriptors. A Kohonen's self-organizing Neural Network (KohNN) map is used to split the whole dataset into a training set including 380 compounds and a test set consisting of 172 compounds. Six selected descriptors from ADRIANA.Code and the six selected descriptors from Cerius~2 are used as the input descriptors for building quantitative models by using Partial Least Square (PLS) analysis and Support Vector Machine (SVM) Regression. Then, another two models are built based on nine descriptors selected by a combination of ADRIANA.Code and Cerius~2 descriptors using PLS and SVM, respectively. For the three SVM models, correlation coefficients (r) of 0.87, 0.89 and 0.88 are achieved; and standard deviations (s) of 10.98, 9.72 and 9.14 are obtained for test set.These ADME models can be applied to predict unknown drug properties and to screen drugs virtually.