Neural network
3 methods · 1995–2020
Neural network: Neural networks outside the deep-learning lineage: the early sequence-determination networks, and small task-specific models that score one local decision, such as an ion-series assignment or a leucine/isoleucine call.
Neural networks outside the deep-learning lineage: the early sequence-determination networks, and small task-specific models that score one local decision, such as an ion-series assignment or a leucine/isoleucine call.
The earliest of its 3 methods is Neural-network CID peptide sequencing (1995); 2 more have followed.
| Methods | 3 |
| Papers describing them | 4 |
| Authors | 6 |
| Active | 1995-10-01 to 2020-08-21 |
| Deep learning | 3 of 3 |
| Kinds | adjacent (2), algorithm |
| Acquisition | DDA (3) |
Methods (3)
Oldest first, by the paper that describes each one.
- Neural-network CID peptide sequencing (1995): Artificial-neural-network approach for peptide sequence determination from high-energy CID spectra.
- b-/y-ion staged neural network (2012): Staged neural network that models ion fragmentation patterns and estimates the posterior probability of each ion type, used to pick the informative peaks out of an MS/MS spectrum before sequencing. The motivation is search-space control: too many peaks and the candidate peptide space grows exponentially, too few and the ion ladder has gaps that can only be explained by permutations of amino acid combinations, so either way candidate quality drops. Reported to beat other preprocessing techniques and to cut the candidate search space substantially without losing candidate quality. The authors note the step matters to any interpretation of MS/MS spectra, de novo or not. Two papers a year apart: the BIBM 2012 classifier, and the two-stage version in Proteome Science 2013 that states it improves on it.
- Leucine/isoleucine discrimination (2020): Deep-neural-network discrimination of the isomeric residues leucine and isoleucine, which are practically indistinguishable in de novo sequencing from ordinary tandem MS data. Rather than relying on the characteristic satellite ions that EThCD fragmentation produces, it learns from raw spectra directly, searching a broader range of the signal for other evidence that separates the two.
Papers describing them (4)
- Peptide sequence determination from high-energy collision-induced dissociation spectra using artificial neural networks (1995, Journal of the American Society for Mass Spectrometry, peer-reviewed)
- A neural network approach to the identification of b-/y-ions in MS/MS spectra (2012, 2012 IEEE International Conference on Bioinformatics and Biomedicine (BIBM), peer-reviewed)
- Identification of b-/y-ions in MS/MS spectra using a two stage neural network (2013, Proteome Science, peer-reviewed)
- Discrimination of Leucine and Isoleucine in De Novo peptide sequencing using deep neural networks (2020, thesis)