Prediction of antihypertensive activity of pyridazinone derivatives through multivariate image analysis applied to QSAR

The antihypertensive activities of a series of pyridazinone derivatives have been modeled using an image-based approach for quantitative structure-activity relationships (MIA-QSAR). First, multivariate image analysis (MIA) descriptors were achieved from bidimensional images. Then, two regression methods were applied to correlate such descriptors with the antihypertensive activities: principal component regression (PCR) and principal component-artificial neural network (PC-ANN) with back-propagation learning algorithm, respectively. Correlation ranking procedure was used to rank the principal components and entered them into the models. The PC-ANN-based model for this series of compounds demonstrated higher predictive ability than the PCR-based model. The results supported that PCR model could predict the antihypertensive activity in pyridazinone derivatives with R (2) a parts per thousand yen 0.726 for training set and R (2) a parts per thousand yen 0.622 for validation set, whereas nonlinear method PC-ANN could predict the antihypertensive activity with R (2) a parts per thousand yen 0.970 for training set and with R (2) a parts per thousand yen 0.997 and R (2) a parts per thousand yen 0.982 for validation and test sets, respectively. The obtained results indicated that MIA descriptors may be useful to predict antihypertensive activity of pyridazinone derivatives.