Peptide ion mobility adds an extra dimension of separation to mass spectrometry-based proteomics. The ability to accurately predict peptide ion mobility would be useful to expedite assay development, or as an additional constraint for peptide identification. Although there are methods to accurately predict peptide mobility in drift tube ion mobility, more work is needed to predict mobility through the high-field asymmetric waveform ion mobility (FAIMS). Here, we show that prediction of peptide FAIMS ion mobility is not a simple regression due to peptides observed at multiple mobilities, but we successfully model this problem as multi-label classification. We trained two separate models, a random forest and a long-term short-term memory neural network. Both models had different strengths, and the ensemble average of model predictions produced higher f2 score than either model alone. Finally, we explore why the models are wrong, and demonstrate predictive performance of f2=0.66 (AUROC = 0.928) on a new test dataset of nearly 40,000 different E. coli peptides.

Peptides were analyzed by DDA with a single FAIMS CV value from -15 to -115. Only -20 to -95 were kept because there were few identifications in more extreme CV values. All identifications were imported into skyline, and peaks were integrated across all FAIMS CV runs to determine the transmission profiles.

Peptides from either human K562 or E. coli proteomes digested with trypsin and LysC.