ScopeNovo reranks de novo peptide sequencing candidates by spectrum-to-spectrum similarity

Summary. Researchers at the City University of Hong Kong reported in Bioinformatics, published online on 15 October, ScopeNovo, a method for de novo peptide sequencing from tandem mass spectra. De novo sequencing is used to characterise unknown proteins and post-translational modifications without a reference database, but deep-learning models usually choose a sequence on decoding score alone, which can favour candidates that the observed spectrum does not directly support. ScopeNovo generates several candidate peptides with an existing model, pi-PrimeNovo, and reranks them by comparing the input spectrum with a representative spectrum for each predicted sequence.

Research context (RUO). On the standard nine-species benchmark datasets V1 and V2, the authors report that ScopeNovo raised the average peptide accuracy of pi-PrimeNovo by 6.1% and 4.0% for well-annotated species, and by 3.4% and 1.9% in a second setting for species that are not well annotated. In the first setting its peptide accuracies were 0.711 and 0.737, against 0.687 and 0.682 for RankNovo, which the authors describe as the best known method; in the second setting the two were comparable. The paper is a computational proteomics methods study with no biological or clinical testing. The code is released in Python under an MIT licence on GitHub, with test data on Zenodo. For mass spectrometry and proteomics laboratories, the contribution is a spectrum-grounded reranking step that can sit on top of existing de novo sequencing models.