Measuring co-localization of ion images
This repository is devoted to a project on measuring co-localization of mass spectrometry images. The project is carried out by the Alexandrov team at EMBL Heidelberg. We created a webapp for ranking pairs of ion images, engaged external experts to rank images from their public data from METASPACE, consolidated the results into a gold standard set of ranked pairs of ion images, and, finally, developed and evaluated various measures of co-localization.
- Katja Ovchinnikova: pixel-based co-localization method development, gold standard preparation
- Alexander Rakhlin: deep learning based co-localization method development
- Lachlan Stuart: development and implementation of the RankColoc web app
- Sergey Nikolenko: PI for the deep learning work
- Theodore Alexandrov: supervision, gold standard preparation
Creating gold standard ion images
Using public METASPACE datasets
We used public datasets from METASPACE, a community-populated knowledge base of metabolite images. Please see the section Acknowledgements acknowledging contributors of the used data.
RankColoc was rapidly prototyped using the METASPACE codebase as a foundation, allowing its back-end, image display and ranking to be reused. The RankColoc-specific changes can be found in this commit range.
The gold standard is available here.
The ion images are available under
gs_imgs2 file names. To join both files into one arhive run
cat gs_imgs* > gs_imgs.tar.gz
The initial expert rankings can be found in
rankings.csv, the filtered gold standard with average rankings is in
Measures requiring no learning
Measures requiring no learning are available in the jupyter notebook
Measures based on deep learning
Measures based on deep learning are available here.
We are planning to integrate the best methods into https://metaspace2020.eu.
Unless specified otherwise in file headers or LICENSE files present in subdirectories, all files in this repository are licensed under the Apache 2.0 license.