The rotationally averaged collision cross-section (CCS)
determined
by ion mobility-mass spectrometry (IM-MS) facilitates the identification
of various biomolecules. Although machine learning (ML) models have
recently emerged as a highly accurate approach for predicting CCS
values, they rely on large data sets from various instruments, calibrants,
and setups, which can introduce additional errors. In this study,
we identified and validated that ion’s polarizability and mass-to-charge
ratio (m/z) have the most significant
predictive power for traveling-wave IM CCS values in relation to other
physicochemical properties of ions. Constructed solely based on these
two physicochemical properties, our CCS prediction approach demonstrated
high accuracy (mean relative error of <3.0%) even when trained
with limited data (15 CCS values). Given its ability to excel with
limited data, our approach harbors immense potential for constructing
a precisely predicted CCS database tailored to each distinct experimental
setup. A Python script for CCS prediction using our approach is freely
available at https://github.com/MSBSiriraj/SVR_CCSPrediction under the GNU
General Public License (GPL) version 3.