The ability to identify peptides, proteins and their associated post-translational modifications using mass spectrometry is directly linked to the fragmentation level of peptide precursor ions. Typically, a well-fragmented precursor ion generates a tandem mass spectrum with a high signal-to-noise ratio, which facilitates peptide sequence identification. Precursor ion properties, such as mass-to-charge ratio (m/z), charge state, and ion mobility coefficient influence the level of collision energy required for optimal fragmentation. Nonetheless, most mass spectrometers do not make use of all these pieces of information when attempting to determine the optimal collision energy for a given peptide. Furthermore, current approaches optimizing precursor ion fragmentation, typically rely on knowing peptide sequences prior to analysis or attempting many fragmentations with different collision energies during the experiment.

To address these limitations, we present a novel supervised learning strategy that uses precursor ion property measurements from mass spectrometry to determine in real-time the collision energy level that optimizes precursor ion fragmentation. Specifically, we designed an artificial neural network using precursor ion properties such as m/z, charge state and ion mobility coefficient to predict precursor ion fragmentation level when a given collision energy is applied. To train our network, we built a dataset of data-dependent acquisition (DDA) mass spectrometry analyses of commercial human cell lysates (Promega) on a timsTOF Pro (Bruker) mass spectrometer using fixed collision energy values in each analysis, ranging from 5 to 100eV. This algorithm is then used to determine in real-time, during mass spectrometry analysis, the optimal collision energy for a given precursor ion.

We show that our algorithm accurately predicts precursor ion relative fragmentation based on the collision energy applied in DDA mass spectrometry analyses of whole cell lysates (r²=0.72). When our tool selects the optimal collision energy level for a given precursor ion, peptide and protein identifications from PEAKS de novo sequencing increase by 15% and 5%, respectively, compared to the instrument"s default collision energy settings (false discovery rate (FDR)<1%). Despite being trained on DDA data, our algorithm can determine the optimal collision energy for peptide ions in each isolation window in real-time during data-independent acquisition (DIA) mass spectrometry experiments and increase the number of peptide identifications by 9.0% (FDR<1%). Finally, we show that when applied to human cell lysates enriched for phosphorylated peptides, our method identified 12% more post-translationally modified peptides (FDR<1%).

Overall, our novel precursor ion fragmentation optimization method improves peptide, protein and post-translational modification identifications, thereby providing a more comprehensive characterization of samples analyzed by mass spectrometry.