On application of constitutional descriptors for merging of quinoxaline data sets using linear statistical methods

The present paper is an attempt for unifying two different quinoxaline data sets with a wide range of substituents in 2, 3, 7, and 8 positions having excellent antitubercular activities with a view to developing robust and reliable structure-activity relationships. The merging has been performed for these two sets of quinoxaline 1,4-di-N-oxides derivatives comprising 29 and 18 compounds, respectively, on the basis of constitutional descriptors, which denotes the structural characterization of the molecules. Principal component analysis was performed to see the distribution of the compounds from two data sets for the constitutional descriptors. The distribution of compounds in score plot based on constitutional descriptors suggests unification of quinoxaline data sets which is useful for the model development. Outlier detection was performed from the standpoint of residual analysis of the partial least squares regression models. The superiority of the constitutional descriptors over other calculated molecular descriptors has been established from the standpoint of leave-one-out cross-validation technique associated with partial least squares regression analysis. Internal validation through the leave-many-out methodology was also performed with good results, assuring the stability of the models. The results obtained from linear partial least squares regression analysis lead to a statistically significant and robust quantitative structure-activity relationship modeling.