CCS Prediction

Rotationally averaged collision cross section (CCS) values are established as a complementary point of identification in targeted screening, alongside m/z, retention time t[R] and product ion spectra. Where standards are available, a common approach is to generate a compound library containing m/z, t[R], CCS, product ion and meta data [1], and use these attributes in a targeted screen or compound search. In the absence of experimental data, predicted values could be used as an alternative, a question frequently asked by customers in the context particularly of semi-targeted and untargeted screening.

At present, there are two dominant approaches to CCS prediction:

  1. A priori prediction using the structure of the molecule as a starting point, generation of multiple conformers, identification of protonation sites, energy minimization, forcefield calculations [2]. This approach is computationally expensive and can take several days per compound to complete. Scaling to use cloud/compute farm resources reduces the time needed per compound but this approach does not lend itself to higher throughput analysis.
  2. Machine learning, in which a set of compounds, each with experimental CCS values, and a set of molecular descriptors, is used to train a model [3,4]. Once constructed, prediction of the CCS value for a new compound is trivial and completed in as short amount of time. While the mathematics behind machine learning is complex, once the model is developed, it is relatively easy to use.


  1. J.P. Williams, D. Eatough, L.A. Gethings, C.J. Hughes, M. Towers, L. Nye, S. Lai, R. Tyldesley-Worster, J.P. Vissers, S. Dhungana, Automatic CCS and MS/MS Library Creation and Application for Large Scale Metabolic Profiling, in: ASMS Proc., 2016: p. MP 258.
  2. C. Lapthorn, F.S. Pullen, B.Z. Chowdhry, P. Wright, G.L. Perkins, Y. Heredia, How useful is molecular modelling in combination with ion mobility mass spectrometry for “small molecule” ion mobility collision cross-sections?, Analyst. (2015). doi:10.1039/c5an00411j.
  3. M. Heinonen, H. Shen, N. Zamboni, J. Rousu, Metabolite identification and molecular fingerprint prediction through machine learning, Bioinformatics. (2012). doi:10.1093/bioinformatics/bts437.
  4. Z. Zhou, X. Shen, J. Tu, Z.J. Zhu, Large-scale prediction of collision cross-section values for metabolites in ion mobility-mass spectrometry, Anal. Chem. (2016). doi:10.1021/acs.analchem.6b03091.

Share this article