Sequence-to-sequence translation from mass spectra to peptides with a transformer model

Abstract

A fundamental challenge for any mass spectrometry-based proteomics experiment is the identification of the peptide that generated each acquired tandem mass spectrum. Although approaches that leverage known peptide sequence databases are widely used and effective for well-characterized model organisms, such methods cannot detect unexpected peptides and can be impractical or impossible to apply in some settings. Thus, the ability to assign peptide sequences to the acquired tandem mass spectra without prior information -- de novo peptide sequencing -- is valuable for gaining biological insights for tasks including antibody sequencing, immunopeptidomics, and metaproteomics. Although many methods have been developed to address this de novo sequencing problem, it remains an outstanding challenge, in part due to the difficulty of modeling the irregular data structure of tandem mass spectra. Here, we describe Casanovo, a machine learning model that uses a transformer neural network architecture to translate the sequence of peaks in a tandem mass spectrum into the sequence of amino acids that comprise the generating peptide. We train a Casanovo model from 30 million labeled spectra and demonstrate that the model outperforms several state-of-the-art methods on a cross-species benchmark dataset. We also develop a version of Casanovo that is fine-tuned for non-enzymatic peptides. Finally, we demonstrate that Casanovo's superior performance improves the analysis of immunopeptideomics and metaproteomics experiments and allows us to delve deeper into the dark proteome.

De novo peptide sequencing with InstaNovo: Accurate, database-free peptide identification for large scale proteomics experiments

“…Both DeepNovo and PointNovo pass their peak encodings to an LSTM or an output layer to predict the next amino acid. Casanovo frames the problem as a sequenceto-sequence problem [32] and employs a transformer encoder-decoder framework to process and predict sequences of amino acids. Point-Novo and Casanovo were both retrained using their respective official GitHub repositories (github.com/volpato30/DeepNovoV2 for PointNovo, and github.com/Noble-Lab/casanovo for Casanovo).…”

Novel Proteoform Discovery by Precise Semi-De Novo Sequencing of Novel Junction Peptides

“…Finally, the SoftMax function is used to convert the array Z 20 to an output array P 20 with values between 0 and 1 (eq 4), which is the probability distribution of the AA category. 1 20 PSM Scoring Function in CNovo.…”

“…Casanovo was recently reported to use a transformer architecture for de novo peptide sequencing with higher accuracy than DeepNovo and Novor. 20 We also compared SpliceNovo with Casanovo. In DS1-JP, DS2-JP, and DS3-JP, the peptide recall rate for the top-1 results of SpliceNovo was 1.9, −2.8, and −0.5% higher than that of Casanovo (Figure S12a−c).…”

Great Gains in Mass Spectrometry Data Interpretation

“…1 Over these decades, numerous algorithmic advances have steadily improved MS data interpretation for sequence identification. 2 The pace of this progress continues unabated along many avenues, including improved interpretation of complex spectra from multiplexed data independent acquisition (plexDIA), 3 de novo sequencing with new embeddings and transformer neural network architectures, 4 improvements in open searches and the identification of peptide modifications, 5 and improved models of isotopic compositions.The latter advances are exemplified by a new approach termed Conditional fragment Ion Distribution Search (CIDS). 6 CIDS can substantially increase sequence identification rates for peptides labeled by using heavy water ( 2 H) or 15 N since such peptides have structural isomers and the distributions of their fragment ions have been difficult to predict.…”

“…Over these decades, numerous algorithmic advances have steadily improved MS data interpretation for sequence identification . The pace of this progress continues unabated along many avenues, including improved interpretation of complex spectra from multiplexed data independent acquisition (plexDIA), de novo sequencing with new embeddings and transformer neural network architectures, improvements in open searches and the identification of peptide modifications, and improved models of isotopic compositions.…”