Effect of coordinate rotation on 3D molecular descriptors computed by DragonX

TL;DR

This paper investigates how coordinate system rotations affect 3D molecular descriptors computed by DragonX, revealing that descriptor values vary with orientation, which impacts their reliability in property prediction models.

Contribution

It identifies the sensitivity of DragonX 3D descriptors to coordinate rotations, highlighting a fundamental issue affecting descriptor reliability and suggesting the need for improved computation methods.

Findings

Descriptor values vary systematically with coordinate rotations

Rotation sensitivity impacts descriptor reliability

Highlights need for improved descriptor computation methods

Abstract

Quantitative structure-property relations (QSPR) employing descriptors derived from the 3D molecular structure are frequently applied for property prediction in various fields of research. In particular, DragonX is one of the most widely used software packages for descriptor calculation. The reliability of 3D molecular descriptors computed by DragonX has lately been investigated, thereby focusing on the effect of computational methods used for molecular structure optimization on the accuracy of the resulting molecular descriptors. The present contribution extends the analysis to a more intrinsic problem of DragonX descriptor evaluation resulting from the sensitivity of the computed 3D descriptors on the coordinate system used for molecule description. Evaluating several 3D descriptors for converged molecular structures rotated around all 3 spatial axes (affine coordinate transformations) yields systematically varying descriptor values. Since this unphysical behavior severely affects the descriptor reliability, the present contribution is meant to summarize these findings for later improvement of either descriptor definitions or the DragonX implementations.