Novel global and local 3D atom-based linear descriptors of the Minkowski distance matrix: theory, diversity-variability analysis and QSPR applications

A new family of alignment-free 3D descriptors based on TOMOCOMD-CARDD framework has been designed, namely 3D-linear indices. In this report, we have proposed the use of a generalized form of the geometric pairwise atom-atom distance matrix as structural information matrix. This matrix, denominated as non-stochastic, uses as matrix form of linear maps as well as their algebraic transformations: stochastic, double stochastic and mutual probabilities matrices. The methodology for 3D-QSAR studies is based on the combined use of global and local approaches. Principal component analysis reveals that the novel indices are capable of capturing structural information not codified by the indices implemented in the DRAGON's software. Moreover, Shannon's entropy based variability analysis comparing the 3D-linear indices with some relevant descriptors suggests that the former encode similar-to-better amount of structural information than these descriptors. Finally, a search for the best regressions for congeneric databases in QSPR modeling was performed. The overall results demonstrates satisfactory behavior.