Global Spectrum Interpretation (GSI): A global, integrative approach for the protein inference problem
Résumé
The computational approach of bottom-up proteomics experiments usually consists of two successive steps: first, matching mass spectra to peptides, second inferring proteins based on the peptide-spectrum matches (PSMs) selected during the first step. Despite recent advances in PSMs rescoring and protein inference using Graph Neural Networks or Bayesian Networks, identifications accuracy and proteome coverage in shotgun proteomics experiments can still be improved. Indeed, several proteins may rely on the same subset of peptides, which make them difficult to discriminate. Besides, some spectra can be hard to identify and match several peptides. In that case, assigning a spectrum to a specific peptide in the first step prevents spectral information to reach the second step, which may lead to erroneous results. We therefore developed GSI, a model that simultaneously assigns spectra to peptides and infers proteins, thanks to a tripartite graph structure. In fact, GSI tries to find the smallest set of proteins best explaining the simultaneously chosen PSMs, thus limiting the impact of the anticipated PSM choice on the protein inference. Moreover, GSI integrates deep-learning computed biological data, namely peptide detectability, to help to distinguish proteins sharing peptides. Experimental results obtained on various mass spectrometry datasets show consistent identification and quantification results compared to well-established models such as X!TandemPipeline or EPIFANY.
| Origine | Fichiers produits par l'(les) auteur(s) |
|---|---|
| Licence |