Application of partial least squares and radial basis function neural networks in multivariate imaging analysis-quantitative structure activity relationship: Study of cyclin dependent kinase 4 inhibitors

The detailed application of multivariate image analysis (MIA) method for the evaluation of quantitative structure activity relationship (QSAR) of some cyclin dependent kinase 4 inhibitors is demonstrated. MIA is a type of data mining methods that is based on data sets obtained from 2D images. The purpose of this study is to construct a relationship between pixels of images of investigated compounds as independent and their bioactivities as a dependent variable. Partial least square (PLS) and principal components-radial basis function neural networks (PC-RBFNNs) were developed to obtain a statistical explanation of the activity of the molecules. The performance of developed models were tested by several validation methods such as external and internal tests and also criteria recommended by Tropsha and Roy. The resulted PLS model had a high statistical quality (R2 = 0.991 and RCV2=0.993) for predicting the activity of the compounds. Because of high correlation between values of predicted and experimental activities, MIA-QSAR proved to be a highly predictive approach.

Graphical abstractThe detailed application of multivariate image analysis method for the evaluation of QSAR of some cyclin dependent kinase 4 inhibitors is demonstrated.Figure optionsDownload full-size imageDownload high-quality image (59 K)Download as PowerPoint slideResearch highlights▶ We obtained stable statistical models in combination with the multivariate image analysis (MIA) for the evaluation of quantitative structure activity relationship (QSAR) of some cyclin dependent kinase 4 inhibitors. ▶ PLS represented superior results and it could predict about 99% of variances in the inhibitory activity data. ▶ MIA-QSAR proved to be a highly predictive approach. ▶ To the best of our knowledge there is not any report on the combination of multivariate image analysis method and radial basis function neural network and PLS as regression methods for building of QSAR models.