Українська  English  Русский  


Inventory reference

ISSN 1812-7231 Klin.inform.telemed. Volume 7, Issue 8, 2011, Pages 88-92


N. Ya. Golovenko, V. E. Kuz'min, A. G. Artemenko, M. A. Kulinsky, P. G. Polishchuk, I. Yu. Borisyuk


A. V. Bogatsky Physico-Chemical Institute NAS Ukraine, Odesa

Article title

Prediction of bioavailability of drugs by the method of classification models

Abstract (resume)

The study was conducted using the methods of classification trees and "random forest" (Random Forest). Sample size was 628 compounds. To assess the bioavailability of classification models were constructed, predicted this option for two (low and acceptable), or three classes (low, medium, high). To construct the QSPR classification models used the method of simplex representation of molecular structure. For each sample separately QSPR models were constructed using "random forest".

Interclassification mistakes for models with three classes of high, hence the predictive ability of this model is low. Regression model also has a low predictive ability (R2oob = 0.294). Simplifying the classification of two classes has a better predictive ability.

Given the intersection of models and varying bioavailability in a certain range of values obtained in the model we have introduced a confidence interval within 10% percent of the border. Thus wereobtained classification models taking into account the confidence interval, which significantly increase the predictive ability.

Thus, the method of classification trees is a promising tool for preliminary analysis of the bioavailability of potential drugs. However, there is significant influence of various physiological factors that reduce the bioavailability of drugs before they enter the systemic circulation, which are difficult to establish the simulation, since it requires additional experimental research. In our opinion, this method well predicts the bioavailability of low molecular weight compounds, absorption of which occurs by simple diffusion.


bioavailability, method of classification trees, method of "random forest"


Full-text version