Deep learning accelerates proteomic research

Date: 27th May 2019

Currently mass spectrometry (MS) is  a fundamental  method used to identify or quantify proteins, however, the outcome relies heavily on database searching or library matching, and interpretation of the data can be time-consuming and requires a great deal of skill.

Now researchers from the Technical University of Munich have addressed this problem through the application of deep learning, with a finding published in Nature Methods. By utilising tandem mass spectrometry, an approach in which molecules are separated by molecular weight by one mass spectrometer, fragmented as they exit, and identified on the basis of their fragments by a second mass spectrometer.  They have trained a deep neural network, to read all parts of the spectra to determine fragment (and hence protein) identification, a significant improvement compared to current evaluation software which is only able to  use partial data.  Coined Prosit, this AI software can lead to an increased identification with a >10× lower false discovery rate.

It is hoped that Prosit, which is freely available and will become a global tool that will transform the proteomic field.  The ultimate goal from a clinical perspective is that it will enable faster, more accurate diagnosis of disease, drug targeting and analysis.  From a basic research perspective, Prosite is also not restricted to humans and can be used more broadly, opening-up new possibilities of discovery in heterologous systems.

Gessulat, S., T. Schmidt, D. P. Zolg, P. Samaras, K. Schnatbaum, J. Zerweck, T. Knaute, J. Rechenberger, B. Delanghe, A. Huhmer, U. Reimer, H.-C. Ehrlich, S. Aiche, B. Kuster and M. Wilhelm (2019). “Prosit: proteome-wide prediction of peptide tandem mass spectra by deep learning.” Nature Methods 16(6): 509-518.