Structural, Physicochemical and Stereochemical Interpretation of QSAR Models Based on Simplex Representation of Molecular Structure

In this chapter we describe different structural, physicochemical and stereochemical approaches towards interpretation of QSAR models based on simplex representation of molecular structure (SiRMS). These techniques are feasible due to the flexible nature of SiRMS, which may encode not only structural and physicochemical features of molecules, but also stereochemical ones (to represent molecules with different types of chirality). The developed approaches to structural and physicochemical interpretation do not depend on used machine learning methods that makes it possible to easily interpret traditional “black box” models like Support Vector Machine and Random Forest. We demonstrated an applicability of the developed interpretation approaches in a number of case studies including classical Hammett and Free-Wilson analysis, as well as several data sets with various physical and biological end-points. A good correspondence of the interpretation results with classical Hammett and Free-Wilson approaches supports validity of the proposed approaches. The analysis of different data sets with different end-points showed three possible scenarios of QSAR models’ interpretation depending on the mechanisms of action for studied compounds that brings us to a conclusion that despite all models are interpretable, not all end-points are. The stereochemical interpretation was applied to the classical Cramer’s set of steroids and to the data set that includes compounds with mixed central and axial chirality. In both cases we demonstrated the substantial contribution of the chiral descriptors in 2.XD QSAR models and revealed certain stereochemical features, which have the biggest contributions to investigated properties. As SiRMS represents an attractive framework for developing predictive and interpretable models, we developed several open-source software tools to make it available for the community. They are discussed at the end of the chapter.