Collision cross section prediction of deprotonated phenolics in a travelling-wave ion mobility spectrometer using molecular descriptors and chemometrics.

The combination of ion mobility and mass spectrometry (MS) affords significant improvements over conventional MS/MS, especially in the characterization of isomeric metabolites due to the differences in their collision cross sections (CCS). Experimentally obtained CCS values are typically matched with theoretical CCS values from Trajectory Method (TM) and/or Projection Approximation (PA) calculations. In this paper, predictive models for CCS of deprotonated phenolics were developed using molecular descriptors and chemometric tools, stepwise multiple linear regression (SMLR), principal components regression (PCR), and partial least squares regression (PLS). A total of 102 molecular descriptors were generated and reduced to 28 after employing a feature selection tool, composed of mass, topological descriptors, Jurs descriptors and shadow indices. Therefore, the generated models considered the effects of mass, 3D conformation and partial charge distribution on CCS, which are the main parameters for either TM or PA (only 3D conformation) calculations. All three techniques yielded highly predictive models for both the training (R(2)SMLR = 0.9911; R(2)PCR = 0.9917; R(2)PLS = 0.9918) and validation datasets (R(2)SMLR = 0.9489; R(2)PCR = 0.9761; R(2)PLS = 0.9760). Also, the high cross validated R(2) values indicate that the generated models are robust and highly predictive (Q(2)SMLR = 0.9859; Q(2)PCR = 0.9748; Q(2)PLS = 0.9760). The predictions were also very comparable to the results from TM calculations using modified mobcal (N2). Most importantly, this method offered a rapid (<10 min) alternative to TM calculations without compromising predictive ability. These methods could therefore be used in routine analysis and could be easily integrated to metabolite identification platforms.