|Tobias Schmidt is a fourth-year Ph.D. candidate at the chair of Proteomics and Bioanalytics of Prof. Kuster in Freising, Technical University Munich, Germany. Before coming to the Technical University of Munich for his MSc in molecular biotechnology he spent three years at the Karlsruhe Institute of Technology studying mathematics. His research interest is in in-memory databases (ProteomicsDB), combining modern machine learning with high quality synthetic and real-world data as well as porting legacy (academic) systems to new technologies. His doctoral thesis work explores among other things the usage of his prediction tools for data analysis pipelines requiring high-quality spectral libraries.
Real-time Spectrum Prediction in Skyline via ProteomicsDB’s gRPC Interface to PrositProsit is able to accurately predict fragment ion intensities and retention times of peptides by deep learning. However, deep-learning requires GPUs for predictions that are not yet readily available in many labs and thus limit its applicability. In order to circumvent this shortcoming, we made Prosit available via gRPC on ProteomicsDB, such that Skyline is able to directly request predictions in real-time and on-demand. Read More
The service went online in December and within two months served over 330 users, covering 17 countries and had an uptime of 100%. Prediction of 5000 spectra takes on average 300 milliseconds. We envisage that Prosit will not be limited to fragment ion intensities and retention times, but will also serve other relevant peptide properties to Skyline. As a proof of principle, we started to develop a new model that predicts the expected charge state distribution of a peptide (R=0.7).