#27 · Proteomics

Deep and adaptive models are refining predictive steps in mass spectrometry workflows

The common thread is data-driven models that set or correct acquisition and interpretation parameters in mass spectrometry workflows rather than relying on fixed rules. msengine adapts mass and retention-time windows for HDX-MS signal extraction, and Desktop iDMS predicts FAIMS/DMS separation and compensation voltages for glycosphingolipid epimer resolution. Bromo, a transformer, ranks peptide precursors by relative MS response with charge state, outperforms sequence-based methods, and can be fine-tuned. ProFormer applies a transformer to tabular MS1 peptide-ion features to classify single-cell and plasma proteomes. The chromatogram deconvolution algorithm separates chimeric DIA MS/MS spectra using temporal correlation with precursor profiles, aiding lipid identification. The dominant thread is thus learned or adaptive prediction applied to new data types. What is new is parameter selection conditioned on the data itself, as in msengine's adaptive windows and iDMS voltage prediction, rather than static settings.