Method for non-autoregressive de novo sequencing of amino acid sequence of peptides
RU2858180C1 · Institute of Biomedical Chemistry
| Patent | RU2858180C1 ↗ |
| Filed | 2025-07-16 |
| Published (granted) | 2026-03-13 |
| Assignee | Institute of Biomedical Chemistry (printed Федеральное государственное бюджетное научное учреждение “Научно-исследовательский институт биомедицинской химии имени В.Н. Ореховича” (ИБМХ)) |
| Claims | PowerNovo2 |
Abstract
FIELD: molecular biology. SUBSTANCE: present invention relates to a method for non-autoregressive de novo sequencing of the amino acid sequence of peptides. The method includes two stages. At the training stage, a posterior distribution of latent variables corresponding to each annotated peptide sequence is modelled using a posterior neural network; a prior distribution of latent variables is formed using a prior flow network; optimising the parameters of the formed and trained model by maximising the variational lower bound on the likelihood. At the prediction stage: converting electronic data of the studied spectrum obtained as a result of tandem mass spectrometric detection of peptides into a vector representation; inputting the vector representation of the converted electronic data into a pre-trained model to obtain corresponding latent variables by sampling from the prior distribution set by the trained prior neural network; predicting the decoded peptide sequence based on the latent variables generated by the trained posterior neural network and the embeddings of the studied spectrum with independent and simultaneous generation of each token of the output sequence by the trained prior neural network; checking the correspondence of the molecular weight of the predicted peptide sequences to the molecular weight of the precursor; correlating the electronic data of each identified peptide with the electronic data of amino acid sequences of real proteins from a pre-created database of known and predicted proteins. EFFECT: implementation of the invention provides increased accuracy and reduced analysis time. 3 cl, 2 dwg, 2 ex
Why it is here
Institute of Biomedical Chemistry (Moscow): non-autoregressive de novo sequencing with a latent-variable model; the inventors are PowerNovo2’s authors.
Inventors (5)
Denis V. Petrovskiy, Kirill S. Nikolsky, Liudmila I. Kulikova, Vladimir R. Rudnev, Anna L. Kaysheva
An inventor links to an author page only where the same person is evident, never on a name alone.
Catalogued papers it cites (1)
- PowerNovo: de novo peptide sequencing via tandem mass spectrometry using an ensemble of transformer and BERT models (2024) added by the examiner
Matched from the patent’s 2 literature citations by the rules the paper citation graph uses.
Cited by (1 patent families, 1 organisations)
- 海南大学 academic: CN122369590A
Every family, of any applicant, that cites this one, as Google Patents lists them; an assignee is given its English name.