Journal of Proteomics & Bioinformatics
Open Access

Methods for Development of De Novo Sequencing Algorithms

Susan Williams^{* *}Correspondence: Susan Williams, Department of Biological Studies, Flinders University, Adelaide, Australia, Email:

Author info »

Description

Development of de novosequencing algorithms

Manual de novo sequencing is time-consuming and laborintensive. The interpretation of spectra is usually done using algorithms or programmes that come with the mass spectrometer. An ancient method is to list all probable peptides for the precursor ion in mass spectrum and compare each candidate's mass spectrum to the experimental spectrum [1]. The peptide with the most comparable spectra has the greatest likelihood of being the correct sequence. However, there could be a great number of potential peptides. A precursor peptide with a molecular weight of 774, for example, has a total of 21,909,046 potential peptides. It takes a long time, even if it is done on a computer.

Another method is "subsequencing," which matches small sequences of peptides that reflect only a portion of the total peptide rather than listing the entire sequence of potential peptides [2]. When sequences that closely match the fragment ions in the experimental spectrum are discovered, they are extended one by one by residues until the best match is identified.

The third method employs a graphical representation of the data, in which fragment ions with similar mass differences of one amino acid residue are linked by lines. It is easier to obtain a clear image of ion series of the same kind in this manner [3]. This method may be useful for manual de novo peptide sequencing; however it is ineffective in a high-throughput environment.

The graph theory is the fourth strategy that is believed to be successful. Bartels was the first to mention using graph theory in de novo peptide sequencing. Peaks in the spectrum become vertices in a graph known as a "spectrum graph." A directed edge will be applied if two vertices have the same mass difference of one or more amino acids. This sort of algorithm includes the SeqMS, Lutefisk, and Sherenga algorithms.

Deep learning approaches have recently been used to overcome the difficulty of de novo peptide sequencing. DeepNovo was the first breakthrough, using a convolutional neural network topology to achieve significant increases in sequence accuracy and entire protein sequence assembly without the use of databases [4]. In order to extract information from a raw spectrum, additional network architectures, such as PointNet, have been used. The sequence prediction problem is then applied to the de novo peptide sequencing problem. Neuralnetwork- based de novo peptide sequencing models will create the most likely next amino acid until the predicted peptide's mass matches the precursor mass, given a previously predicted partial peptide sequence.

Conclusion

PepNovo is a high-throughput de novo peptide sequencing tool that scores peptides using a probabilistic network. One spectrum interpretation usually takes less than 0.2 seconds. PepNovo outperforms Sherenga, PEAKS, and Lutefisk, amongst other popular algorithms. PepNovo+, a new version, is now available. DeepNovo outperformed earlier approaches, such as PEAKS, Novor, and PepNovo, by a large margin, according to the benchmark analysis in the original publication. The Tensorflow framework is used to implement DeepNovo in Python.

References

1. Webb-Robertson BJ, Cannon WR. Current trends in computational inference from mass spectrometry-based proteomics. Brief Bioinform. 2007;8(5):304-17.

   [Crossref] [Google Scholar] [PubMed]

2. Smith LM, Kelleher NL. Proteoform: A single term describing protein complexity. Nat Methods. 2013;10(3):186-7.

   [Crossref] [Google Scholar] [PubMed]

3. Capriotti AL, Cavaliere C, Foglia P, Samperi R, Laganà A. Intact protein separation by chromatographic and/or electrophoretic techniques for top-down proteomics. J Chromatogr A. 2011;1218(49):8760-76.

   [Crossref] [Google Scholar] [PubMed]

4. Roth MJ, Plymire DA, Chang AN, Kim J, Maresh EM, Larson SE, et al. Sensitive and reproducible intact mass analysis of complex protein mixtures with superficially porous capillary reversed-phase liquid chromatography mass spectrometry. Anal Chem. 2011;83(24):9586-92.

   [Crossref] [Google Scholar] [PubMed]