If you're doing metabolomics and need to identify unknown compounds from mass spectra, this handles the grunt work of loading data from mzML, MGF, or MSP files, cleaning it up with 40+ filters, and running similarity searches against spectral libraries. The modified cosine similarity accounts for precursor mass differences, which matters for fragment matching. It's built for reproducible pipelines, so you can chain filters together and process entire libraries consistently. Note this is for small molecule metabolomics, if you're doing proteomics LC-MS/MS work, you want pyopenms instead. The SpectrumProcessor class makes it easy to standardize everything before similarity scoring.
npx skills add https://github.com/k-dense-ai/scientific-agent-skills --skill matchms