Abstract

HUICI CORRALES, Adonis et al. Predicting the affinity of A_2A adenosine receptor antagonists using decision trees. Biot. Veg. [online]. 2019, vol.19, n.2, pp. 113-126.  Epub June 01, 2019. ISSN 2074-8647.

Neurodegenerative diseases are being treated by modulating adenosine receptors with more effective, safe and selective antagonists. The objective of the study was to develop a methodology to obtain classification models based on decision tree algorithms and descriptors from 0D to 2D of non-congenital families of organic compounds to qualitatively predict ligand-RAA2A affinity. For this purpose, a non-congeneric database of 315 antagonists was constructed and cured with its inhibition constant in nano molar, labeled as potent and weak. The Dragon and ISIDA / QSPR programs were used to calculate molecular descriptors and five groups of descriptors were obtained. In each group 50 descriptors were selected using the mRMR criterion. The database was divided into Training, Test and External series through a random selection and a generalized k-means cluster analysis. Classifiers were developed and validated using the WEKA program. The results were analyzed using the statistical tests of Friedman and Wilcoxon. The significant influence of parameter m of algorithm J48 on the predictivity was verified for the models that used the descriptors of the aug.a-b and hyb.aug.a groups of ISIDA / QSPR. The best performance model was obtained from the selected descriptors of the ISIDA-all group with a value of m = 6 and reached 90.6% prediction on the External series. The methodology developed to obtain classification models based on decision tree algorithms and descriptors from 0D to 2D of non-congenital families of organic compounds is effective in qualitatively predicting ligand-RAA2A affinity with accuracy, specificity and selectivity greater than 90 %.

Keywords : classification; machine learning; modeling; QSAR.

        · abstract in Spanish     · text in Spanish     · Spanish ( pdf )