Masserstein: robust linear deconvolution by optimal transport
Ciach, M. A.; Miasojedow, B.; Skoraczynski, G.; Majewski, S.; Startek, M.; Valkenborg, D.; Gambin, A.
Show abstract
A common theme in many applications of computational mass spectrometry is fitting a linear combination of reference spectra to an experimental one in order to estimate the quantities of different ions, potentially with overlapping isotopic envelopes. In this work, we study this procedure in an abstract setting, in order to develop new approaches applicable to a diverse range of experiments. We introduce an application of a new spectral dissimilarity measure, known in other fields as the Wasserstein or the Earth Movers distance, in order to overcome the sensitivity of ordinary linear regression to measurement inaccuracies. Usinga a data set of 200 mass spectra, we demonstrate that our approach is capable of accurate estimation of ion proportions without extensive pre-processing required for state-of-the-art methods. The conclusions are further substantiated using data sets simulated in a way that mimics most of the measurement inaccuracies occurring in real experiments. We have implemented our methods in a Python 3 package, freely available at https://github.com/mciach/masserstein.
Matching journals
The top 5 journals account for 50% of the predicted probability mass.
Similar papers in this journal
Similar papers in this journal
- Extremely fast and accurate open modification spectral library searching of high-resolution mass spectra using feature hashing and graphics processing units 96%
- Quality control for the target decoy approach for peptide identification 95%
- Improving peptide-level mass spectrometry analysis via double competition 95%
Similar papers in this journal
Similar papers in this journal
- Common data models to streamline metabolomics processing and annotation, and implementation in a Python pipeline 94%
- Spec2Vec: Improved mass spectral similarity scoring through learning of structural relationships 94%
- Multienzyme deep learning models improve peptide de novo sequencing by mass spectrometry proteomics 94%
Similar papers in this journal
"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.