Revealing the differences in collision cross section values of small organic molecules acquired by different instrumental designs and prediction models

The number of open access databases containing experimental and predicted collision cross section (CCS) values is rising and leads to their increased use for compound identification. However, the reproducibility of reference values with different instrumental designs and the comparison between predicted and experimental CCS values is still under evaluation. This study compared experimental CCS values of 56 small molecules (Contaminants of Emerging Concern) acquired by both drift tube (DT) and travelling wave (TW) ion mobility mass spectrometry (IM-MS). The TWIM-MS included two instrumental designs (Synapt G2 and VION). The experimental (CCSN2)-C-TW values obtained by the TWIM-MS systems showed absolute percent errors (APEs) < 2% in comparison to experimental DTIMS data, indicating a good correlation between the datasets. Furthermore, (CCSN2)-C-TW values of [M - H](- )ions presented the lowest APEs. An influence of the compound class on APEs was observed. The applicability of prediction models based on artificial neural networks (ANN) and multivariate adaptive regression splines (MARS), both built using TWIM-MS data, was investigated for the first time for the prediction of (CCSN2)-C-DT values. For [M+H]+ and [M - H](- )ions, the 95th percentile confidence intervals of observed APEs were comparable to values reported for both models indicating a good applicability for DTIMS predictions. For the prediction of (CCSN2)-C-DT values of [M+Na](+) ions, the MARS based model provided the best results with 73.9% of the ions showing APEs below the threshold reported for [M+Na](+). Finally, recommendations for database transfer and applications of prediction models for future DTIMS studies are made.

Automated summary

The number of open access databases containing experimental and predicted collision cross section (CCS) values is increasing, which is important for compound identification. However, the reproducibility of reference values with different instrumental designs and the comparison between predicted and experimental CCS values is still being evaluated. This study compared experimental CCS values of 56 small molecules acquired by both drift tube (DT) and travelling wave (TW) ion mobility mass spectrometry (IM-MS). The results showed good correlation between the datasets of the two methods. Prediction models based on artificial neural networks (ANN) and multivariate adaptive regression splines (MARS) were also found to be applicable for the prediction of CCS values. Recommendations for database transfer and applications of prediction models are provided for future studies.