In mass spectrometry, even the best experiments produce complex data, so “garbage in, garbage out” is always lurking in the background. So we spend our days sorting signals from noise with a smile.
DecoyGen
In MS-based proteomics, false discovery rates are typically estimated with target–decoy strategies, but common decoy generators (e.g., reversing or shuffling sequences) often produce unrealistic peptides and distort score distributions, undermining FDR confidence. We use protein language models to generate more realistic decoy peptides by masking and resampling amino acids, and validate them with target–decoy separability tests and MS-based shotgun proteomics data (DDA and DIA).
IgNovo
A Python-Rust package for de novo protein and peptide sequencing from mass spectrometry data using combination of deep learning and graph-based methods.