A de novo peptide sequencing model based on non-autoregressive Transformer and dual learning
CN120089196A · Liaoning University · priority 2025
| Patent | CN120089196A ↗ |
| Status | Pending |
| Priority date | 2025-01-26 |
| Filed | 2025-01-26 |
| Published | 2025-06-03 |
| Assignee | Liaoning University |
Abstract
A peptide de novo sequencing model based on non-autoregressive transformers and dual learning. The method comprises the steps of firstly dividing original mass spectrum data into different resolutions through multi-scale processing, secondly extracting potential characteristics of the original mass spectrum by using a self-attention mechanism of a transducer encoder, thirdly, deducing peptide sequences in parallel by using a non-autoregressive transducer decoder based on the potential characteristics of the original mass spectrum and peptide precursor information, and fourthly, reconstructing mass spectrum data according to the generated peptide sequences through the transducer encoder by using a dual-learning paradigm. The training objective of this model is to minimize the negative log likelihood loss and mass spectral reconstruction loss of the peptide sequence. Compared with the prior art, the method can better capture global and local characteristics in mass spectrum data, obviously improve the prediction accuracy of mass spectrum to peptide sequences, and simultaneously improve the reasoning speed of a model. The model is suitable for large-scale mass spectrum data analysis, and has high-efficiency calculation and good generalization performance.
Why it is here
Liaoning University: non-autoregressive Transformer de novo sequencing with dual learning.
Inventors (4)
刘宏生, 马健烨, 刘汀, 张力
An inventor links to an author page only where the same person is evident, never on a name alone.