AntoinePassemiers / Wasserstein-Mass-Spectrometry

Unofficial implementation of the paper "The Wasserstein Distance as a Dissimilarity Measure for Mass Spectra with Application to Spectral Deconvolution"

Wasserstein Mass Spectrometry

This software is a non-official implementation of the paper "The Wasserstein Distance as a Dissimilarity Measure for Mass Spectra with Application to Spectral Deconvolution" by S. Majewski, M. A. Ciach, M. Startek, W. Niemyska, B. Miasojedow and A. Gambin.

Default solver is based on a problem-specific genetic algorithm. This allows for numerical stability and high speed.

Figure: t-distributed Stochastic Neighbor Embedding of MS1 ESI-QTOF spectra with precomputed matrix Wasserstein distances. The colormaps gives the mass of each molecule (point) expressed in Daltons.

How to use it

Run mass spectral deconvolution algorithm on an empirical spectrum and a folder containing the theoretical spectra. Paths to theoretical spectra should be specified in molecule_list.txt. A concrete example of dataset is provided in example-msd/ folder.

deconvms path/to/empirical/spectrum.txt path/to/molecule_list.txt path/to/molecules/folder --niter 50000

Compute Wasserstein dissimilarity between two mass spectra:

wassms path/to/first/record/file.txt path/to/second/record/file.txt --m W

Compute Euclidean distance between two mass spectra:

wassms path/to/first/record/file.txt path/to/second/record/file.txt --m E

Installation

Compile the project with g++:

make release

This will create the two executable files wassms and deconvms.

Dependencies

The project is built on top of Eigen, whose path is assumed to be stored in a Eigen3_DIR variable.