2010 Volume 9 Issue 4 Pages 167-176
In the present work, we investigated a practical use of the Topological Fragment Spectra based support vector machines (SVMs) for drug discovery by virtual screening of antihypertensive drugs. We employed two MDDR databases which were released in different years (2001 and 2003). First, we developed a classification model, which consists of collective SVMs, to identify antihypertensive drugs. We used a structure database, MDDR 2001 to develop and validate the SVM models. The data involved 9503 antihypertensive drugs and the negatives of 66521 compounds. The obtained SVM models gave good performance in the learning, in the validation and in the prediction. To evaluate the ability of the classifiers in practical use, we also prepared an external test set that was derived from the set of compounds in MDDR 2003, but not from the set in 2001 (relative complement). The external test set consisted of 19387 compounds, 396 antihypertensive drugs and 18991 negative compounds. The computational trial of the prediction with the external test set successfully identified 158 compounds of the 369 antihypertensive drugs. The results indicate that the present approach can be useful in virtual screening for drug discovery.
