Spectral Prediction Features as a Solution for the Search Space Size Problem in Proteogenomics
Steven Verbruggen, Siegfried Gessulat, Ralf Gabriels, Anna Matsaroki, Hendrik Van de Voorde, Bernhard Küster, Sven Degroeve, Lennart Martens, Wim Van Criekinge, Mathias Wilhelm, Gerben Menschaert
Abstract
Proteogenomics suffers from statistical issues as the sequencing information inflates the database size. To compensate for this, rescoring with the machine learning-based spectrum predictors MS 2 PIP and Prosit was implemented in a proteogenomics approach. This was demonstrated for both ribosome profiling and nanopore RNA-Seq derived databases. Postprocessing with Percolator showed that these techniques result in recovered and often even elevated stringency levels and identification rates. In this way, it allows to validate novel proteoforms through proteogenomics with unsurpassed confidence levels.