Highlight talk – Theme: Data
Abstract
Speaker: Juho Rousu Identification of small molecules from biological samples remains a major bottleneck in understanding the inner working of organisms and their ecosystems. In metabolomics, vast majority of the molecules in tandem mass spectra remain unknown. In recent years, major advances have been gained through adopting machine learning models trained on large databases of reference spectra. I will present CSI:FingerID (http://www.csi-fingerid.org), a software that identifies small molecules from their tandem mass spectra. CSI:FingerID is currently the most accurate methods for identification of metabolite from metabolomics samples, providing 150% more correct identifications than the closest competing method. In the presentation, I will give an overview of the technology underlying CSI:FingerID, combining multiple kernel learning, fragmentation tree models and arrays of support vector machines. I will also give a peek to the recent developments that allow three orders of magnitude faster training and also provide improved identification accuracy.
Authors
Kai Dührkop, Friedrich-Schiller-University Jena, Germany
Huibin Shen, Aalto University, Finland
Marvin Meusel, Friedrich-Schiller-University Jena, Germany
Céline Brouard, Aalto university, Finland
Sebastian Böcker, Friedrich Schiller University Jena, Germany
Juho Rousu, Aalto University, Finland
Source of publication
2015, Proceedings of the National Academy of Sciences, vol 112 no 41, pages 12580-12585.