Inclusion of conformational information in MIA-QSPR

Abstract

MIA-QSPR (Multivariate Image Analysis applied to Quantitative Structure-Property Relationships) is an efficient technique for molecular modeling developed with the aim of simplifying quantitative analyses that seek correlating structural features of chemical compounds with physical-chemical/biological properties (QSPR studies). Essentially, it relies on the idea that, for a congeneric series of molecule images, the changing in substituent groups will generate variations in the pixel orientations (positions), which, in turn, can explain the visual changing observed in the response variable values (physical-chemical/biological property). The molecular descriptors employed in this technique are obtained from bidimentional (2D) projections of the chemical structures under analysis. Since MIA-QSPR has a 2D approach, intuitively, it does not include spatial information (3D), which can be considered a limitation of this technique, once it is well-known that spatial features rule a significative part of the physical-chemical/biological molecular behavior. However, regarding MIA-QSPR, there are no evidences in literature that the inclusion of this type of information would, indeed, improve the prediction ability of QSPR models. In this sense, this work aimed to develop a strategy for inserting conformational and bioconformational information into the MIA-QSPR descriptors, and a detailed analysis of the results of such inclusion. For means of comparison, this work was divided into three parts: i) initially, a traditional MIA-QSPR model was built for a class of compounds with a physical-chemical property independent of a biological receptor (logKoc); ii) a second MIA-QSPR model was built using bidimentional projections resulted from the fully structural optimization of a set of molecules in a receptor free environment; iii) finally, for a third set of compounds, the most likely bioconformations were determined through molecular docking technique and, based on the bidimentional projection of these structures, the last MIA-QSPR model was built. It is worth mentioning that the second molecule set has a biological target, but, until currently, this receptor is not completely known, then a bioconformational study was not required; on another hand, for the last set of compounds, the biological target is already fully elucidated, which supports the molecular docking step performed herein. As a result, it was verified that the MIA-QSPR technique is capable of encoding conformational information; however, including this type of information resulted in an imperfect alignment of the congeneric substructures, which is crucial for building efficient MIA-QSPR prediction models. Accordingly, the models obtained from the optimized/docked structures generated less satisfactory calibration models than the conventional ones, and, therefore, one can conclude that the atomic connectivity and properties (e.g. van der Waals radii and electronegativity) are more efficient molecular descriptors for MIA-QSPR technique than spatial parameters considered herein.