AbstractMass spectrometry-based proteomics allows to study all proteins of a sample on a molecular level. The ever increasing complexity and amount of proteomics MS-data requires powerful and yet efficient computational and statistical analysis. In particular, most recent bottom-up MS-based proteomics studies consider either a diverse pool of post-translational modifications, employ large databases – as in metaproteomics or proteogenomics, contain multiple isoforms of proteins, include unspecific cleavage sites or even combinations thereof and thus face a computationally challenging situation regarding protein identification. In order to cope with resulting large search spaces, we present a deep learning approach that jointly embeds MS/MS spectra and peptides into the same vector space such that embeddings can be compared easily and interchangeable by using euclidean distances. In contrast to existing spectrum embedding techniques, ours are learned jointly with their respective peptides and thus remain meaningful. By visualizing the learned manifold of both spectrum and peptide embeddings in correspondence to their physicochemical properties our approach becomes easily interpretable. At the same time, our joint embeddings blur the lines between spectra and protein sequences, providing a powerful framework for peptide identification. In particular, we build an open search, which allows to search multiple ten-thousands of spectra against millions of peptides within seconds. yHydra achieves identification rates that are compatible with MSFragger. Due to the open search, delta masses are assigned to each identification which allows to unrestrictedly characterize post-translational modifications. Meaningful joint embeddings allow for faster open searches and generally make downstream analysis efficient and convenient for example for integration with other omics types.Availabilityupon [email protected]
Case Report: Locking Plate for Cubitus Varus Correction in a 7-Year-Old Girl With Osteogenesis Imperfecta
